1/2026: Ai For Clinical Decision Support

Ai For Clinical Decision Support

Presentation number: PP 01

Abstract number: 7-ISABS-2026

INTEGRATING 3D SIMULATION AND PSYCHOLOGICAL ASSESSMENT IN POST-BARIATRIC BREAST RESHAPING: A PROSPECTIVE CONTROLLED STUDY IN A PUBLIC HEALTH SETTING

Bassetto Franco¹, Vindigni Vincenzo¹, Agaraj Patris¹, Riccardi Sveva¹

¹Padova University, Padova, Italy

franco.bassetto@unipd.it; vincenzo.vindigni@unipd.it

Post-bariatric women frequently develop severe breast deformities following massive weight loss, with significant physical discomfort and psychosocial distress. Optimizing preoperative counseling is crucial to align expectations and improve postoperative satisfaction. The aim of this prospective controlled study is to evaluate the clinical and psychological impact of integrating three-dimensional (3D) simulation into preoperative counseling for breast reshaping within a public healthcare system. One hundred consecutive post-bariatric patients undergoing mastopexy, reduction mammaplasty, or mastopexy with or without implants are enrolled at a university-affiliated tertiary referral center and randomized 4:1 to standard counseling or counseling enhanced with 3D simulation. Baseline evaluation includes comprehensive clinical assessment, anthropometric and surgical data collection, and validated psychological and body image questionnaires. Primary outcome is postoperative satisfaction at 6 weeks. Secondary outcomes include expectation–outcome alignment, decisional conflict reduction, changes in body image perception, rate of surgical plan modifications after counseling, treatment postponement or refusal, and perceived correspondence between simulated and surgical results. Preliminary findings indicate improved expectation realism, greater patient engagement, and reduced decisional uncertainty in the 3D simulation group, without increased operative time or complication rates. The integration of structured 3D simulation into preoperative counseling may represent a valuable tool to enhance psychological preparedness, shared decision-making, and early satisfaction in post-bariatric breast reshaping within a public health framework.

Keywords: breast, mastopexy, mwl, counseling, prothesis

Presentation number: PP 02

Abstract number: 66-ISABS-2026

AI DECODING DCIS: IMAGING SIGNATURES ON CONTRAST-ENHANCED MAMMOGRAPHY

Bojanić Kristina¹,², Galić Petra³, Galić Dora², Kralik Kristina⁴, Steiner Justinija¹, Ivanac Gordana⁵,⁶, Smolić Robert²,⁷, Smolić Martina²

¹Health Center Osijek-Baranja County, Osijek, Croatia, ²Faculty of Dental Medicine and Health Osijek, University of Osijek, Osijek, Croatia, ³University Hospital Centre Osijek, Osijek, Croatia
⁴Faculty of Medicine Osijek, University of Osijek, Osijek, Croatia, ⁵School of Medicine, University of Zagreb, Zagreb, Croatia, ⁶University Hospital Dubrava, Zagreb, Croatia, ⁷Mursa Medical Center, Osijek, Croatia

bojanic.kristina@gmail.com

The aim of this study was to assess the potential of contrast-enhanced mammography (CEM) combined with artificial intelligence (AI) for non-invasive characterization of tumor biology, with a particular focus on ductal carcinoma in situ (DCIS) as a distinct entity. In this retrospective single-center study, 654 women underwent breast imaging and histopathological evaluation, with 113 lesions analyzed, including 10 cases of DCIS (9%). Imaging features on CEM, AI-derived malignancy scores (iCAD ProFound AI®), and pathological characteristics were assessed. Compared with invasive cancers, DCIS lesions demonstrated significantly lower conspicuity (p<0.001), more frequent association with suspicious microcalcifications (86% vs 36–57%, p=0.02), and a higher prevalence of partial enhancement (57% vs 5%, p=0.005). Most DCIS lesions did not present as a distinct mass, corresponding to a non-mass imaging phenotype. No DCIS cases showed axillary lymph node involvement, in contrast to invasive cancers. AI-derived malignancy scores were lower in DCIS compared to invasive tumors, without statistical significance. These findings indicate that DCIS exhibits a specific imaging signature characterized by low conspicuity, limited enhancement, and strong association with microcalcifications. The integration of functional imaging and AI provides measurable signals that may reflect tumor behavior beyond conventional assessment. This approach supports imaging-based phenotyping and highlights the potential for non-invasive risk stratification and personalized management in early breast cancer.

Keywords: breast cancer, ductal carcinoma in situ, contrast-enhanced mammography, artificial intelligence, personalized medicine

Presentation number: PP 03

Abstract number: 64-ISABS-2026

FROM PIXELS TO PHENOTYPE: CAN AI-ENHANCED MAMMOGRAPHY DECODE BREAST CANCER BIOLOGY?

Bojanić Kristina^1,2, Galic Petra³, Galic Dora², Kralik Kristina⁴, Steiner Justinija¹, Ivanac Gordana^5,6, Smolic Robert², Smolić Martina²

¹Health Center Osijek-Baranja County, Osijek, Croatia; ²Faculty of Dental Medicine and Health Osijek, University of Osijek, Osijek, Croatia; ³Department of Plastic, reconstructive and aesthetic surgery, University Hospital Centre Osijek, Osijek, Croatia; ⁴Faculty of Medicine Osijek, Josip Juraj Strossmayer University of Osijek, Osijek, Croatia; ⁵School of Medicine, University of Zagreb, Zagreb, Croatia; ⁶University Hospital Dubrava, Zagreb, Croatia

bojanic.kristina@gmail.com

The aim of this study was to evaluate whether tumor biology can be inferred non-invasively using contrast-enhanced mammography (CEM) and artificial intelligence (AI). This retrospective single-center study included 399 patients with abnormal screening findings (BI-RADS 0), with 113 analyzed lesions, including 76 malignant and 37 benign lesions. All lesions were assessed using CEM and AI-based malignancy scoring (iCAD ProFound AI®). Tumors were classified according to molecular subtypes and grouped as luminal versus HER2-positive and triple-negative. CEM demonstrated subtype-related imaging phenotypes, with luminal tumors more frequently appearing as irregular, spiculated masses with heterogeneous enhancement, while non-luminal tumors more often presented as round lesions with rim or homogeneous enhancement. AI analysis showed significantly higher malignancy scores in malignant compared to benign lesions (70.5% vs 38.0%, p < 0.001), with good diagnostic performance (AUC = 0.744, sensitivity 71%, specificity 70%). Although AI scores varied across molecular subtypes, these differences were not statistically significant. These findings suggest that while tumor biology is not fully captured by conventional morphology, the integration of functional imaging and AI moves imaging beyond detection toward biological characterization and supports the development of non-invasive biomarkers in precision breast cancer diagnostics.

Keywords: breast cancer, contrast-enhanced mammography, artificial intelligence, molecular subtypes, precision medicine

Presentation number: PP 04

Abstract number: 15-ISABS-2026

EGFR-AI: A MACHINE LEARNING MODEL FOR PREDICTING EARLY POSTOPERATIVE RENAL FUNCTION

Brenner Eva¹, Bulić Luka^1,2,3, Vrbanović Mijatović Vilena^4,5

¹St. Catherine Specialty Hospital, Zagreb, Croatia; ²Algebra Bernays University, Zagreb, Croatia; ³School of Medicine, Josip Juraj Strossmayer University of Osijek, Osijek, Croatia; ⁴Department of Anesthesiology, Reanimatology and Intensive Care Medicine, University Hospital Center Zagreb, Zagreb, Croatia; ⁵School of Medicine, University of Zagreb, Zagreb, Croatia

eva.brenner13@gmail.com

Acute kidney injury is a frequent and serious postoperative complication, making early identification of patients at risk essential for optimizing clinical management. This study aimed to develop a machine learning–based predictive model capable of estimating renal function in the early postoperative period using routinely collected perioperative data. A retrospective cohort of 200 surgical patients admitted to the intensive care unit of the University Hospital Centre Zagreb between January and July 2024 was analyzed. Preoperative and intraoperative variables were used to predict postoperative estimated glomerular filtration rate (eGFR), calculated using the CKD-EPI equation and categorized into three classes (G1, G2, and G3+). A two-layer machine learning architecture (“eGFR-AI”) combining XGBoost classifiers and logistic regression was developed to predict postoperative eGFR categories. The final model achieved an accuracy of 0.75 and ROC-AUC of 0.92. Feature importance analysis identified chronic kidney disease, arterial hypertension, age, and sepsis or shock as key predictors of postoperative renal function. These results demonstrate that machine learning models based on routinely available clinical data may support early risk stratification and personalized postoperative management of surgical patients.

Keywords: postoperative kidney function, machine learning, artificial intelligence, glomerular filtration rate, personalized postoperative care

Presentation number: PP 05

Abstract number: 102-ISABS-2026

DIAGNOSTICS OF LUPUS ANTICOAGULANT IN COAGULATION LABORATORY

Đurek Valentina¹

¹Clinical Hospital Centre Zagreb, Zagreb, Croatia

vale.durek@gmail.com

Lupus anticoagulant is a heterogeneous group of antibodies that interfere with phospholipid-dependent coagulation pathways in vitro, resulting in prolongation of APTV, PV or dRVVT (dilute Russell vipervenom time). The purpose of testing lupus anticoagulant is in suspected primary or secondary antiphospholipid syndrome, together with aCL and anti-β2-GPI, in elderly patients with deep vein thrombosis or embolism and in case of repeated miscarriages and suspected thrombophilia. At the Clinical Hospital Center Zagreb, Laboratory of Hematology and Coagulation performs lupus anticoagulant test on a daily basis. Blood is collected in a coagulation tube with 3.2% sodium citrate anticoagulant. From the plasma sample first are performed screening tests: diluted APTV and dRVVT (diluted Russell Viper Venom Test, LA1) and if necessary, confirmatory tests: LA2 ​​Confirmation Reagent and Lupus anticoagulant test (immuno test). For reliable interpretation of LA test results, samples should not be taken from patients under anticoagulant therapy. In 2025, 1000 samples were processed, of which 134 (13.4%) tested positive for lupus anticoagulant and 788 (78.8%) were negative. Part of the patients, 78 (7.8%) of them were on direct oral anticoagulant therapy and such samples are not processed in the confirmatory tests. Lupus anticoagulant is estimated to be present in 2 to 4% of the general population, but true prevalence is unclear. Among the different types of antiphospholipid antibodies, the lupus anticoagulant antibodies, characterized by their interference with clotting assays with low phospholipid content, show the strongest association with both thromboembolic and obstetric complications.

Keywords: lupus anticoagulant, coagulation, APTV, antiphospholipid, plasma

Presentation number: PP 06

Abstract number: 38-ISABS-2026

MULTI-BAND REHABILITATION OF TDCS AT BAIHUI (GV20) FOR ABNORMAL POST-STROKE NEURAL SYNCHRONIZATION: A PHASE-AMPLITUDE JOINT VALIDATION

Zhou Junwei¹, Wei De², Gao Yueming¹

¹College of Physics and Information Engineering, Fuzhou University, Fuzhou City, China; ²Fuzhou University Affiliated Provincial Hospital, Fuzhou City, China

fzugym@gmail.com

This study investigates the promoting effect of transcranial direct current stimulation (tDCS) at the Bai​hui acupoint (GV20) on the restoration of brain functional networks in stroke patients and its regulatory mechanism on abnormal synchronization. Through a randomized controlled trial, phase lag index (PLI) was employed to analyze multi-band brain networks on 14 subjects. Combined with weighted phase lag index (wPLI) and amplitude envelope correlation (AEC) methods, a dual-dimensional evaluation system for phase synchronization and amplitude coupling was constructed. The results showed that the experimental group exhibited a significant increase in whole-brain PLI values post-intervention (p=0.016), indicating enhanced global phase synchronization. wPLI analysis revealed a 12.6% increase in gamma-band synchronization, while theta-band synchronization was significantly suppressed by 36.9%, demonstrating its ability to correct abnormal low-​frequency hypersynchronization post-stroke. AEC analysis further indicated that tDCS induced a 26.3% suppression of amplitude synchronization in the theta band, with varying degrees of reduction in the alpha and gamma bands. Graph theory analysis showed that the brain network tended to​ward decentralization, sparsity, and higher efficiency, supporting the transition from pathological coupling to functional integration in neural networks. The study confirms that tDCS at Baihui can effectively improve pathological brain network synchronization post-stroke by modulating cross- frequency neural oscillation patterns. The mechanism involves restoring phase coordination in impaired brain regions and bidirectionally regulating abnormal synchronization across different frequency bands.

Keywords: tDCS, Baihui (GV20), stroke rehabilitation, brain network synchronization, phase-amplitude coupling

Presentation number: PP 07

Abstract number: 21-ISABS-2026

SARCOPENIA AND MENOPAUSE

Kovačić Jelena^1,2, Ivkošić Filip¹, Horvat Magda¹, Ivkošić Ante Branko¹, Jelinčić Petar¹, Miškić Blaženka^3,5, Erceg Ivkošić Ivana^3,4

¹School of Medicine, University of Zagreb, Zagreb, Croatia; ²St. Catherine Specialty Hospital, Zagreb, Croatia; ³Center for Women’s Health, St. Catherine Specialty Hospital, Zagreb, Croatia; ⁴Faculty of Dental Medicine and Health, Josip Juraj Strossmayer University of Osijek, Osijek, Croatia; ⁵Department of Internal Medicine, General Hospital “Dr. Josip Benčevic”, Slavonski Brod, Croatia

kovaciccjelena@gmail.com

The goal of this review was to synthesize current evidence on the impact that menopause has on sarcopenia and to explore artificial intelligence (AI) driven approaches for early detection alongside risk stratification. Sarcopenia, defined most commonly as age-related loss of muscle mass and function, is a major concern in menopause stemming from hormonal shifts – particularly oestrogen decline as well as changes in testosterone, IGF-1 with a rise in inflammatory markers. Menopause is clinically defined as a cessation of menses for over 12 months consecutively. Materials and methods used include an integrative analysis of observational studies, clinical trials along with meta-analyses on menopausal women – later combined with AI-based evaluations of clinical, functional and laboratory parameters from multiple datasets to identify patterns associated with muscle loss. Results consistently show that postmenopausal women exhibit reduced muscle mass, strength, and functional performance compared to premenopausal counterparts, whereas lifestyle factors such as physical activity, protein intake, and resistance training significantly mitigate risk. AI models, by integrating multidimensional data, demonstrate high accuracy in predicting sarcopenia thus allowing early identification of high-risk individuals. To conclude, menopause displays a critical window for sarcopenia onset, thereby by combining traditional clinical assessment tests with AI analytics, a newfound promising strategy offers for more personalized interventions. They must include changes in exercise, nutrition – with special emphasis on superior protein intake and hormonal considerations, as to preserve muscle function and improve long-term health outcomes.

Keywords: AI, menopause, proteins, sarcopenia, women

Presentation number: PP 08

Abstract number: 22-ISABS-2026

LIBIDO IN MENOPAUSE

Kovačić Jelena^1,2, Ivkošić Filip¹, Ivanović Tomislav¹, Ivkošić Ante Branko¹, Ćosić Vesna^2,4, Erceg Ivkošić Ivana^2,3

¹School of Medicine, University of Zagreb, Zagreb, Croatia; ²St. Catherine Specialty Hospital, Zagreb, Croatia; ³Faculty of Dental Medicine and Health, Josip Juraj Strossmayer University of Osijek, Osijek, Croatia; ⁴Poliklinika Ćosić, Slavonski Brod, Croatia

kovaciccjelena@gmail.com

The goal of this summary is to identify influences and explore changes in libido during menopause. Menopausal transition can last a few months or years, typically starting in mid-40s/50s with an age average of 51,3 years. In every individual’s life sexual activity plays an important role. Materials and methods used include artificial intelligence (AI) assessment in hormonal change, lifestyle and relationship dynamics then applied to detect patterns as well as predictive contributors towards sexual dysfunction. Women’s sex lives can be subjected to numerous changes that arise due to hormonal transitions, psychosocial factors alongside cultural influences. Main symptoms that women report in menopause are vaginal dryness, dyspareunia, stunted libido and difficulties achieving an orgasm. Results show that approximately 62% of women report noticeable changes, whereas reduced oestrogen and testosterone values, stress, sleep depletion with relational issues prove to be strongly associated. AI modelling highlights that the combination of hormonal decline and psychosocial factors has the highest predictive value for decreased libido. Treatment options are standard and innovative. Standard option is a hormone replacement therapy in the form of oestrogen, progesterone and/or testosterone. Oestrogen is most often used as a low-dose topical therapy for vaginal dryness thereby reducing discomfort during intercourse, although systemic application also shows improvement. Innovative therapies may also include psychological treatment options since libido is not only a physiological problem, but has psychological, sociocultural and intrapersonal relationships influence as well. Female sexual dysfunction (FSD) requires a holistic approach. Ultimately, menopause-related reduction in libido is multifactorial and integrating AI- assisted assessments with individualized interventions addressing both biological and psychosocial contributors can support sexual wellbeing and improve quality of life.

Keywords: AI, FSD, hormone replacement therapy, libido, menopause

Presentation number: PP 09

Abstract number: 23-ISABS-2026

TESTOSTERONE’S ROLE IN WOMEN’S LIVES

Kovačić Jelena^1,2, Ivkošić Ante Branko¹, Kovačić Đurđica¹, Ivkošić Filip¹, Fureš Rajko^4,5, Dulibić Mario¹

¹School of Medicine, University of Zagreb, Zagreb, Croatia; ²St. Catherine Specialty Hospital, Zagreb, Croatia; ³Center for Women’s Health, St. Catherine Specialty Hospital, Zagreb, Croatia; ⁴Faculty of Dental Medicine and Health, Josip Juraj Strossmayer University of Osijek, Osijek, Croatia; ⁵Department of Gynecology and Obstetrics, Zabok General Hospital and Croatian Veterans Hospital, Zabok, Croatia

kovaciccjelena@gmail.com

This study was made to review the role that testosterone has across women’s lifespan as well as to explore the contribution of artificial intelligence (AI) in advancing research and clinical decision-making regarding female androgen health. A narrative review of latest scientific literature, randomized controlled trials, observational studies and international clinical guidelines was performed focusing on female androgen physiology, age-related hormonal changes and implications of testosterone deficiency. Additionally, the application of AI-driven analytic tools for large-scale biomedical data interpretation along knowledge synthesis was considered. Testosterone contributes towards sexual function, mood regulation, cognitive performance, maintenance of bone turnover and muscle mass. Its levels peak during early reproductive years and progressively decline with ageing, with a more pronounced decrease after menopause or bilateral oophorectomy. In polycystic ovary syndrome testosterone is greatly elevated thereby producing symptoms such as seborrhoea, alopecia, hirsutism, etc. On the other hand, reduced androgen levels are associated with decreased libido, fatigue, inferior wellbeing and changes in body composition. Among clinical indications, the strongest evidence for testosterone therapy is the treatment of hypoactive sexual desire disorder, quite often seen in (peri)menopausal women – randomized controlled trials demonstrated improvements in both libido and overall sexual satisfaction. AI-based approaches may support the identification of complex hormonal patterns, integration of multi-source clinical data and development of predictive models that could enable more personalized therapeutic strategies. Testosterone’s role represents an important yet often underestimated factor in women’s health hence why the integration of AI-assisted analytical frameworks may enrich understanding of female androgen physiology and facilitate advanced precision medicine approaches.

Keywords: AI, menopause, polycystic ovary syndrome, testosterone, women

Presentation number: PP 10

Abstract number: 84-ISABS-2026

FUNCTIONALLY DISTINCT MONOCYTE PHENOTYPES PROPAGATE PSORIATIC ARTHRITIS AND ANKYLOSING SPONDYLITIS

Krešić Ivo¹, Jurleta Nikola^1,6, Planinić Pavao¹, Aničić Sara^3,4, Radošević Marta^3,4, Priselac Sara^4,5, Balen Tomislav^4,5, Zrinski Petrović Katerina⁴, Grčević Danka^3,4, Kelava Tomislav^3,4, Kovačić Nataša^4,5, Ikić Matijašević Marina², Ćavar Ivan¹, Šućur Alan^3,4

¹Department of Physiology, School of Medicine, University of Mostar, Mostar, Bosnia and Herzegovina; ²”Sveti Duh” University Hospital, Zagreb, Croatia; ³Department of Physiology, School of Medicine, University of Zagreb, Zagreb, Croatia; ⁴Laboratory for Molecular Immunology, Croatian Institute for Brain Research, School of Medicine, University of Zagreb, Zagreb, Croatia; ⁵Department of Anatomy, School of Medicine University of Zagreb, Zagreb, Croatia; ⁶Internal medicine clinic with dialysis center, University Hospital Mostar, Mostar, Bosnia and Herzegovina

ivo.kresic@mef.sum.ba

Monocytes play a crucial role in the pathogenesis of spondyloarthropathies (SpA), such as ankylosing spondylitis (AS) and psoriatic arthritis (PsA), by differentiating into key inflammatory cells. Among these, monocyte-derived dendritic cells (moDCs) are of particular interest due to their close interaction with T- cells via the IL-12-23/IL-17 axis. This study aims to analyze the expression of DC markers on blood monocytes, evaluate changes upon their differentiation into DCs, and test their subsequent functionality. PBMCs were isolated from patients with active PsA, AS, and healthy controls. Using flow cytometry, we analyzed a panel of markers for monocyte lineage (CD14, CD16), DC differentiation (CD1a, CD1c, CD141, CD206, CD209), antigen presentation (CD40, MHCII), and costimulation (CD80, CD86). Sorted classical monocytes were cultured with DC-inducing factors, IL-17, or both; a control group underwent spontaneous differentiation. Phagocytic ability was tested via pHrodo assay, and antigen-presenting capacity was assessed in a moDC:T-cell co-culture by measuring cytokine production. Fresh PsA monocytes expressed higher levels of CD206, CD209 and CD1a. In vitro, spontaneously differentiated AS monocytes showed lower CD141 and higher CD40. AS-derived moDCs had heightened phagocytosis but produced less IL-12/23, resulting in lower T-cell IFN-γ and TNF-α production. Conversely, PsA-derived moDCs showed higher expression of costimulatory (CD86) and dendritic (CD1c) markers. Across all groups, IL-17 supplementation increased CD209 while decreasing CD141 and CD1a expression. Conclusion: Our findings reveal distinct pathological pathways in SpA. In psoriatic arthritis, monocytes and moDCs show a phenotype skewed towards heightened antigen presentation. In contrast, cells in ankylosing spondylitis display a profile characterized by enhanced phagocytosis but a reduced capacity to stimulate T-cell responses, suggesting different primary mechanisms of immune dysfunction.

Keywords: autoimmune diseases, peripheral blood monocytes, phenotype, phagocytosis, antigen presentation

Presentation number: PP 11

Abstract number: 17-ISABS-2026

WHEN A DESMOID IS NOT A DESMOID: NODULAR FASCIITIS REVEALED BY MOLECULAR DIAGNOSTICS

Krišto Anđela¹, Skejić Lucija¹, Bulić Luka^2,3,4, Brlek Petar^3,2,5, Primorac Dragan^{1,2,3,6,7,8,9,10,11,12,13}

¹University of Split, School of Medicine, Split, Croatia; ²St Catherine Specialty Hospital, Zagreb, Croatia; ³School of Medicine, Josip Juraj Strossmayer University of Osijek, Osijek, Croatia; ⁴Algebra Bernays University, Zagreb, Croatia; ⁵Department of Molecular Biology, Faculty of Science, University of Zagreb, Zagreb, Croatia; ⁶International Center for Applied Biological Research, Zagreb, Croatia; ⁷Faculty of Dental Medicine and Health, Josip Juraj Strossmayer University of Osijek, Osijek, Croatia; ⁸Gandhinagar Campus, National Forensic Sciences University, Gandhinagar, India; ⁹Eberly College of Science, The Pennsylvania State University, State College, PA, United States of America; ¹⁰The Henry C. Lee College of Criminal Justice and Forensic Sciences, University of New Haven, New Haven, CT, United States of America; ¹¹Sana Kliniken Oberfranken, Coburg, Germany; ¹²School of Medicine, University of Rijeka, Rijeka, Croatia; ¹³School of Medicine, University of Pittsburgh, Pittsburgh, PA, United States of America

andela.kristo01@gmail.com

Nodular fasciitis is a benign self-limited myofibroblastic/fibroblastic tumor of uncertain etiology. It is often misdiagnosed as a malignant tumor due to its rapid growth and increased mitotic activity. Aggressive fibromatosis/desmoid tumor is a rare fibroblastic disease belonging to low-grade malignant tumors since they show high potential for infiltrating surrounding tissues. Due to high recurrence rate, these tumors require strict follow-up. Given the histological overlap between both entities, molecular diagnostics play a crucial role in establishing the diagnosis and course of treatment. A 42-year-old patient presented with recently worsened long-standing limb pain and tingling. Upon admission, magnetic resonance imaging (MRI) of the cervical spine revealed an expansive lesion at the C1-C2 level. The patient underwent surgical treatment, after which she reported symptoms regressing. Initial histopathological and immunohistochemical testing corresponded to a primary mesenchymal neoplasm of the aggressive fibromatosis type. However, further molecular genetic analysis revealed a MYH9::USP6 gene fusion, a characteristic molecular feature of nodular fasciitis that supports the distinction of this benign, self-limited lesion from other mesenchymal neoplasms, including aggressive fibromatosis. Accordingly, the biological nature of the tumor was reinterpreted as a more benign, self-limited proliferation that can morphologically mimic a desmoid-type tumor. This finding is clinically significant because it substantially changes the therapeutic approach, preventing unnecessary aggressive treatments. This case highlights the role of molecular diagnostics in refining and, in some cases, redefining the initial diagnosis. Identification of specific genetic alterations can influence clinical decision-making and guide appropriate therapeutic management therefore preventing unnecessary aggressive treatment while ensuring that patients receive appropriate and individualized care.

Keywords: desmoid tumors, molecular tumor profiling, precision oncology, next-generation sequencing, whole exome sequencing

Presentation number: PP 12

Abstract number: 59-ISABS-2026

AI-DRIVEN HRD DETECTION USING SHALLOW WGS: A COMPARATIVE STUDY FROM UNIVERSITY HOSPITAL OF SPLIT

Kunac Nenad¹, Galeković Mia¹, Anđelinović Šimun¹, Vuko Arijana¹

¹University Hospital of Split, Split, Croatia

nkunac@kbsplit.hr

Homologous recombination deficiency (HRD) is a key biomarker for predicting response to PARP inhibitor therapy in precision oncology. While BRCA1/2 mutations represent the canonical mechanism of HRD, many tumors exhibit HRD without detectable mutations. The goal of this report was to evaluate AI-driven detection of HRD using shallow whole -genome sequencing (sWGS) in a real -world clinical setting. Two tumor samples were analyzed using sWGS combined with targeted BRCA1/2 panel sequencing. Genomic instability was quantified using large genomic alterations (LGA) and loss of parental copy (LPC), and an ensemble of 100 machine learning classifiers was applied to estimate genomic instability probability (GI), with HRD defined as GI > 0.5. Both samples were classified as HRD-positive. The BRCA-mutated sample showed GI 0.84, with dominant structural alterations (LGA 28.96, LPC 1) and a pathogenic BRCA2 mutation. The BRCA-wild-type sample showed higher GI 0.99, characterized by extensive copy number loss (LGA 20.83, LPC 22) without detectable pathogenic variants. Model dispersion was lower in the BRCA-wild-type case, indicating higher prediction stability. These results demonstrate that HRD represents a genomic phenotype detectable independently of BRCA mutation status. AI-driven analysis of sWGS data enables robust identification of HRD across diverse molecular contexts and supports improved patient stratification for targeted therapies in routine clinical practice.

Keywords: HRD, shallow WGS, artificial intelligence, genomic instability, precision oncology

Presentation number: PP 13

Abstract number: 112-ISABS-2026

EFFECTS OF GLP-1 AND GLP-1/GIP AGONIST THERAPY ON IGG N-GLYCOSYLATION IN OVERWEIGHT INDIVIDUALS

Lalić Dora¹

¹GlycanAge d.o.o., Zagreb, Croatia

dora@glycanage.com

Protein glycosylation is a key post-translational modification that modulates protein function. Immunoglobulin G (IgG) N-glycans are established biomarkers, with specific patterns associated with immune function and inflammatory processes. These patterns can be integrated into GlycanAge, a composite metric of biological age reflecting immune system status and response to interventions. GLP-1 and GIP receptor agonists, including semaglutide (Wegovy) and tirzepatide (Mounjaro), are widely used for obesity treatment and have demonstrated systemic metabolic and anti-inflammatory effects. However, their impact on IgG glycosylation and GlycanAge remains insufficiently characterized. Twenty-four individuals with obesity undergoing GLP-1/GIP-based therapy were included. Plasma samples were collected at baseline, 3, and 6 months. IgG N-glycans were enzymatically released, labelled, and analysed using capillary gel electrophoresis with fluorescence detection (CGE-LIF). A total of 27 glycan peaks were quantified and used to derive glycan traits, including galactosylation, sialylation, and bisection. Biological age was calculated using the GlycanAge algorithm. Treatment was associated with a shift in IgG glycosylation towards a less inflammatory profile, with increased anti-inflammatory and reduced pro-inflammatory glycan features. This was accompanied by a decrease in GlycanAge, indicating improved biological age, viewed through the lens of the immune system. However, individual responses varied, with some participants showing minimal or adverse changes. GLP-1/GIP-based therapies are associated with favourable changes in IgG glycosylation and biological age, consistent with reduced chronic inflammation. IgG glycan profiling may serve as a sensitive biomarker for monitoring treatment effects, while highlighting interindividual variability.

Keywords: IgG glycosylation, glycanage, GLP-1 receptor agonists, obesity, biological age

Presentation number: PP 14

Abstract number: 94-ISABS-2026

ASSOCIATION OF IL-1Β RS16944 POLYMORPHISM WITH COAGULATION PARAMETERS IN COVID-19 PATIENTS

Meseldžić Neven¹, Prnjavorac Besim², Dujić Tanja¹, Malenica Maja¹, Glamočlija Una¹, Marjanović Damir^3,4, Bego Tamer¹

¹University of Sarajevo – Faculty of Pharmacy, Sarajevo, Zmaja od Bosne 8, Bosnia and Herzegovina; ²General Hospital Tešanj, Tešanj, Bosnia and Herzegovina; ³Institute for Anthropological Research Zagreb, Zagreb, Croatia; ⁴University of Rijeka – Faculty of Biotechnology and Drug Development, Rijeka, Croatia

neven.meseldzic@ffsa.unsa.ba

COVID-19 is frequently associated with coagulation abnormalities that contribute to disease progression and clinical outcomes. The aim of this study was to investigate the association of IL-1β polymorphism (rs16944) with coagulation parameters and clinical severity in COVID-19 patients. The study included 750 patients from the Bosnian population, stratified into three groups according to disease severity. Coagulation parameters, including prothrombin time (PT), activated partial thromboplastin time (aPTT), platelet count, international normalized ratio (INR), and D-dimer, were analyzed using standard IFCC protocols. Statistical analysis was performed using linear regression models to assess differences between genotypes, while logistic regression adjusted for age and sex was used to evaluate the association with disease severity. In patients with mild clinical presentation, significant differences were observed in coagulation parameters, with AA genotype carriers showing higher PT (p=0.010) and aPTT (p=0.030) values compared to GG and GA genotypes. In patients with moderate disease, significant differences were found in platelet count (p=0.046), with higher values observed in AA genotype carriers. No statistically significant associations were observed in patients with severe clinical presentation. Logistic regression analysis showed no significant association between rs16944 polymorphism and disease severity. These findings suggest that the IL-1β rs16944 polymorphism influences coagulation pathways, particularly in earlier stages of COVID-19, but is not an independent predictor of disease severity. The results highlight the importance of genetic variability in modulating host response and support further investigation of IL-1β as a potential biomarker in clinical risk assessment.

Keywords: COVID-19, disease severity, polymorphisms, coagulation, biomarkers

Presentation number: PP 15

Abstract number: 106-ISABS-2026

MACHINE LEARNING EXPLORATION OF MATERNAL CORTISOL VARIABILITY ACROSS ENVIRONMENTAL AND BIOLOGICAL DETERMINANTS IN PREGNANCY

Mrdjen-Hodžić Rafaela¹, Blažević Sofia Ana², Havaš Auguštin Dubravka¹, Šarac Jelena^1,3, Marjanović Damir^1,4,3, Novokmet Natalija¹

¹Center for Applied Bioanthropology, Institute for Anthropological Research, Zagreb, Croatia; ²Faculty of Science, Department of Biology, University of Zagreb, Zagreb, Croatia; ³Faculty of Biotechnology and Drug Development, University of Rijeka, Rijeka, Croatia; ⁴International Burch University, Sarajevo, Bosnia and Herzegovina

rafaela@inantro.hr

Cortisol, a primary biomarker of the hypothalamic–pituitary–adrenal axis, reflects maternal physiological adaptation to pregnancy-related stress. Increasing evidence suggests that cortisol regulation is influenced by complex interactions between environmental and biological factors, highlighting the relevance of integrative approaches. This study aimed to investigate differences in maternal cortisol levels according to geographic environment (island vs mainland), parity (primipara vs multipara), body mass index (BMI), fetal sex. A cross-sectional analysis on a sample of 337 pregnant women was therefore performed, applying standard statistical methods and exploratory machine learning models to evaluate group differences, potential nonlinear relationships and relative importance of predictors. Results indicated modest variability in cortisol levels across geographic and parity groups with BMI showing a weak association with cortisol variability. Machine learning models suggested that combinations of biological and environmental factors contributed more to cortisol variation than individual predictors; however, the overall explained variance remained low. These findings suggest that maternal cortisol regulation in the second trimester is shaped by multifactorial interactions, while the examined variables alone have limited predictive value. Machine learning approaches may provide a useful framework for exploring complex physiological patterns during pregnancy, although their contribution to individualized risk prediction in this context appears modest.

Keywords: cortisol, pregnancy, machine learning, integrative approach, multifactorial interactions

Presentation number: PP 16

Abstract number: 41-ISABS-2026

INTELLIGENCE-DRIVEN PRECISION SONOTHROMBOLYSIS: ADVANCED DEVICE DESIGN AND PREDICTIVE RBC REPAIR

Pan Yunfan^1,2,4,5, Zhou Junwei¹, Yang Shuang¹, Huang Yan³, Wei De³, Liao Xiangwen², Gao Yueming^1,2

¹College of Physics and Information Engineering, Fuzhou University, Fuzhou, China; ²Interdisciplinary Institute for Medical Engineering, Fuzhou University, Fuzhou, China; ³Fujian Provincial Hospital, Fuzhou University, Fuzhou, China; ⁴Department of Mechanical Engineering, Tsinghua University, Beijing, China; ⁵Harvard Medical School, Boston, Massachusetts, United States of America

panyf@fzu.edu.cn; fzugym@gmail.com

Thrombotic diseases are a leading cause of global mortality. Traditional interventions face limitations such as prolonged duration and vascular injury. This study advances precision medicine by developing an intelligent dual-frequency sonothrombolysis system and ultrasound-responsive nanocarriers, while addressing red blood cell (RBC) hemolysis through predictive modeling.(1) Intelligent Interventional System: A dual-frequency ultrasound device was engineered for targeted thrombosis treatment. Utilizing a microfluidic thrombus chip and extracorporeal platforms, we performed data-driven optimization of parameters (frequency, drug concentration). Animal experiments validated its superior efficacy, providing a foundation for intelligence-powered clinical decision support in interventional therapy. (2) Precision Drug Delivery: To enhance penetration, we developed ultrasound-responsive carriers featuring phase-change nanodroplet-coated microbubbles. Fluorescence imaging elucidated the cavitation-mediated thrombolytic mechanism. This approach aligns with intelligence-enhanced drug development, ensuring high-precision delivery and safety through real-time blood flow monitoring. (3) Predictive RBC Repair Model: To mitigate hemolysis, we investigated RBC fatigue under cyclic stress. A microfluidic device revealed morphological transitions, leading to the establishment of an empirical fatigue model. Furthermore, a regenerative strategy using glucose/oxygen-carrying nanodroplets was proposed. By simulating cytoskeletal extension, a critical threshold model for RBC repair was established, offering a theoretical framework for protecting blood cells during therapy. This research integrates intelligent hardware with predictive biological modeling, providing a comprehensive solution for precision sonothrombolysis. It offers a robust technical foundation for the design of future intelligence-optimized medical devices and regenerative therapies.

Keywords: Intelligence-driven precision sonothrombolysis; Advanced interventional device; Precision drug delivery; Predictive model of RBCs; Microfluidics

Presentation number: PP 17

Abstract number: 5-ISABS-2026

AI-POWERED DETECTION OF FOOD CONTAMINANTS: A PREVENTIVE APPROACH WITHIN THE PRECISION MEDICINE FRAMEWORK – HEAVY METAL DETECTION IN FISH AND MARINE ORGANISMS

Slišković Livia¹, Brnić Sanja², Gregurinčić Adrian², Vukadin Andrea², Žučko Korina²

¹Department of forensic genetics, biology and chemistry, Faculty of forensic sciences, University of Split, Split, Croatia; ²Faculty of forensic sciences, University of Split, Split, Croatia

lsliskovic@forenzika.unist.hr

Food contamination with heavy metals represents a significant challenge for food safety and poses serious risks to human health, particularly through the consumption of fish and marine organisms. Traditional analytical methods for heavy metal detection are often time-consuming, costly, and require extensive sample preparation, highlighting the need for faster and more efficient screening approaches. In this context, artificial intelligence (AI) has emerged as a promising tool for improving contaminant detection and supporting preventive food safety strategies.This review evaluates current AI-based approaches for the detection of heavy metal contamination in fish and marine organisms, with a focus on their application in food safety monitoring. Various machine learning and deep learning models, combined with non-destructive analytical techniques such as hyperspectral imaging, have been applied to classify contaminated and non-contaminated samples of marine organisms and to predict heavy metal concentrations based on different types of input data. The reviewed studies demonstrate that AI-driven methods achieve high accuracy, sensitivity, and rapid data processing, often outperforming conventional analytical and statistical approaches. Deep learning models have proven to be excellent for rapid, non-destructive detection, while machine learning algorithms have demonstrated reliable performance even with limited datasets. Overall, AI-powered detection systems represent an effective complementary tool for enhancing heavy metal detection and monitoring in food production systems. Although further efforts are required for standardization and the collection of reference data, AI-based detection systems have strong potential to contribute to early detection and prevention of exposure, thereby enhancing food safety and supporting both public and individual human health within a preventive and precision medicine framework.

Keywords: artificial intelligence, heavy metals, food contamination, food safety, precision medicine

Presentation number: PP 18

Abstract number: 6-ISABS-2026

AI-BASED PREDICTIVE TOXICOLOGY FOR FOOD ADDITIVES: TOWARDS PRECISION RISK ASSESSMENT IN NUTRITION AND HEALTH- APPLICATION TO ARTIFICIAL SWEETENERS

Slišković Livia¹, Gašparovski Lorena², Marelić Hana², Mandić Paola²

¹Department of forensic genetics, biology and chemistry, Faculty of forensic sciences, University of Split, Split, Croatia; ²Faculty of forensic sciences, University of Split, Split, Croatia

lsliskovic@forenzika.unist.hr

Artificial food additives, particularly sweeteners, are widely used in the food industry, and their safety assessment represents an important challenge in food toxicology. Traditional toxicological evaluation is time-consuming and resource-intensive, highlighting the need for alternative approaches that enable earlier and more efficient risk assessment. In this context, artificial intelligence (AI) has emerged as a powerful tool in predictive toxicology.This review explores the application of AI-based predictive toxicology approaches for food additive safety assessment, with a particular focus on artificial sweeteners, including aspartame, sorbitol and xylitol as a case example. Different computational strategies, such as quantitative structure–activity relationship (QSAR) models and machine learning algorithms, have been applied to predict toxicological endpoints based on molecular descriptors and available experimental data. The reviewed studies indicate that AI-driven models can achieve high predictive accuracy for several toxicity endpoints, including genotoxic and other biologically relevant effects, supporting their potential use in early hazard identification and prioritization of substances for further testing. Evidence from the literature further suggests that AI-supported analyses are effective in screening artificial sweeteners and identifying potential risks under different exposure scenarios, while also highlighting limitations related to data availability, model interpretability, standardization, and the need for experimental validation. Overall, AI-based predictive toxicology represents a promising complementary approach in food safety assessment, contributing to preventive risk management and supporting the broader goals of precision and preventive medicine.

Keywords: artificial intelligence, predictive toxicology, food additives, artificial sweeteners, food safety

Presentation number: PP 19

Abstract number: 52-ISABS-2026

EXPLAINABLE AI IN PRECISION MEDICINE FOR DIAGNOSING AND TREATING RARE DISEASES

Talukder Asoke¹, Haas Roland¹, Sehgal Ajai²

¹Applied Research in Artificial Intelligence, Bangalore, India; ²IKS Health, Seattle, WA, United States of America

asoke.talukder@araisolutions.ai

GenAI (Generative AI) shows great promise in precision medicine, but hallucination and high computational demands remain a problem. We present a novel Neuro-Symbolic AI approach, which utilizes GenAI, Knowledge Graphs (KG), and GraphRAG (Graph Retrieval-Augmented Generation) to address these challenges. We used PubMed, FAERS (FDA Adverse Event Reporting System), UMLS (Unified Medical Language System), NORD (National Organization for Rare Disorders), NCIt (National Cancer Institute Thesaurus), and GO (Gene Ontology) as our knowledge sources. We used GenAI to mine PubMed for 18 entity types and 24 relationships. We mined FAERS for Drug-Drug interactions and Disease-Drug contraindications. We constructed a knowledge graph (KG) with all these entities and relationships and added NCIt and GO into it. We embedded the NORD and UMLS knowledge source as GraphRAG. Natural language user input is passed through named entity recognition (NER) and named entity normalization (NEN) to give us UMLS concept IDs (CUIs). These CUIs are used to fetch the immediate neighbours in the biomedical KG and supplementary knowledge from RAG. The results are passed through fine-tuned SLM GenAI for human understandable language. The integration of Knowledge Graphs and GraphRAG enables AI systems to become explainable. The hallucinations of AI have reduced substantially, reasoning is explainable, accuracy is improved and the system can run on desktop. We do not have sufficient data to quantify these results yet. Neuro-Symbolic AI with integrated GraphRAG offers explainability and reduced hallucination with higher accuracy, critical features that are required for medical applications of AI.

Keywords: neuro-symbolic AI, GraphRAG, GenAI, explainability

Presentation number: PP 20

Abstract number: 66-ISABS-2026

AI DECODING DCIS: IMAGING SIGNATURES ON CONTRAST-ENHANCED MAMMOGRAPHY

Bojanić Kristina^1,2, Galić Petra³, Galić Dora², Kralik Kristina⁴, Steiner Justinija¹, Ivanac Gordana^5,6, Smolić Robert^2,7, Smolić Martina²

¹Health Center Osijek-Baranja County, Osijek, Croatia; ²Faculty of Dental Medicine and Health Osijek, University of Osijek, Osijek, Croatia; ³University Hospital Centre Osijek, Osijek, Croatia; ⁴Faculty of Medicine Osijek, University of Osijek, Osijek, Croatia; ⁵School of Medicine, University of Zagreb, Zagreb, Croatia; ⁶University Hospital Dubrava, Zagreb, Croatia; ⁷Mursa Medical Center, Osijek, Croatia

bojanic.kristina@gmail.com

The aim of this study was to assess the potential of contrast-enhanced mammography (CEM) combined with artificial intelligence (AI) for non-invasive characterization of tumor biology, with a particular focus on ductal carcinoma in situ (DCIS) as a distinct entity. In this retrospective single-center study, 654 women underwent breast imaging and histopathological evaluation, with 113 lesions analyzed, including 10 cases of DCIS (9%). Imaging features on CEM, AI-derived malignancy scores (iCAD ProFound AI®), and pathological characteristics were assessed. Compared with invasive cancers, DCIS lesions demonstrated significantly lower conspicuity (p<0.001), more frequent association with suspicious microcalcifications (86% vs 36–57%, p=0.02), and a higher prevalence of partial enhancement (57% vs 5%, p=0.005). Most DCIS lesions did not present as a distinct mass, corresponding to a non-mass imaging phenotype. No DCIS cases showed axillary lymph node involvement, in contrast to invasive cancers. AI-derived malignancy scores were lower in DCIS compared to invasive tumors, without statistical significance. These findings indicate that DCIS exhibits a specific imaging signature characterized by low conspicuity, limited enhancement, and strong association with microcalcifications. The integration of functional imaging and AI provides measurable signals that may reflect tumor behavior beyond conventional assessment. This approach supports imaging-based phenotyping and highlights the potential for non-invasive risk stratification and personalized management in early breast cancer.

Keywords: breast cancer, ductal carcinoma in situ, contrast-enhanced mammography, artificial intelligence, personalised medicine

Presentation number: PP 21

Abstract number: 75-ISABS-2026

NOCTURNAL HYPOCALCEMIC SEIZURES AS THE FIRST MANIFESTATION OF PSEUDOHYPOPARATHYROIDISM TYPE IA IN A YOUNG CHILD

Vilibić Mirko¹, Kovačić Đurđica¹, Vrbanc Lucija¹, Vinković Maja², Braovac Duje², Krnić Nevena^1,2

¹School of Medicine, University of Zagreb, Zagreb, Croatia; ²Department for Pediatric Endocrinology and Diabetes, University Hospital Zagreb, Zagreb, Croatia

mirko.vilibic@gmail.com

Pseudohypoparathyroidism type Ia (PHP-Ia) results from maternal inactivating GNAS mutations on chromosome 20q13.3 that impair the stimulatory G protein (Gsα) signaling. Tissue-specific paternal imprinting leads to markedly reduced Gsα expression in renal proximal tubules, thyroid and pituitary, resulting in PTH resistance with hypocalcemia and hyperphosphatemia. Patients also exhibit Albright hereditary osteodystrophy (AHO) features including brachydactyly, round face, short stature and ectopic ossifications. We report a case of hypocalcemic seizures in a pediatric patient diagnosed with PHP-Ia, emphasizing diagnostic challenges of epileptic events in children. A 5-year-old boy presented with nocturnal tonic-clonic seizures preceded by hand/foot/face spasms over 3 weeks. History revealed infantile diarrhea, speech delay and previous subcutaneous osteoma removal. The patient presented with macrocephaly (+2.85Zcore), stocky habitus (height +1.20Zscore, weight +1.75Zscore), round face, ataxia and subcutaneous osteoma. Laboratory findings included hypocalcemia (Ca2+ 0.56 mmol/L, Ca 1.33 mmol/L), elevated PTH (60.89 pmol/L), hyperphosphatemia (3.16 mmol/L) and low vitamin D levels. Brain CT revealed symmetric basal ganglia calcifications. The diagnosis of PHP-Ia was established based on clinical and laboratory criteria. Treatment with calcitriol and calcium carbonate normalized calcium and improved neurological findings (spasms and ataxia). With its evolving phenotype, the PHP-Ia diagnosis relies on a combination of clinical and biochemical findings. While molecular testing is definitive, the complex GNAS locus inheritance requires a multi-modal approach including DNA sequencing, methylation studies and copy number variant analysis. Hypocalcemic seizures, as the initial manifestation of PHP-Ia, necessitates differentiation from other causes of epilepsy. Therefore, low calcium should prompt immediate PTH screening for timely detection of this rare but intricate disease.

Keywords: albright hereditary osteodystrophy, basal ganglia calcification, hypocalcemia, pseudohypoparathyroidism, seizures

Presentation number: PP 22

Abstract number: 39-ISABS-2026

A FLEXIBLE CONFORMAL NEAR-FIELD RESONANT SENSING INTERFACE FOR CONTINUOUS MONITORING OF PERIPHERAL ARTERIAL HEMODYNAMICS

Yang Shuang¹, Huang Yan², Gao Yueming¹

¹Fuzhou University, Fuzhou, China; ²Fuzhou University Affiliated Provincial Hospital, Fuzhou, China

241110035@fzu.edu.cn; fzugym@gmail.com

The early detection and proactive management of cardiovascular diseases depend critically on the long-term, continuous monitoring of peripheral arterial hemodynamics. However, conventional non-invasive modalities, such as photoplethysmography (PPG), are fundamentally limited by shallow optical penetration, which constrains their ability to reliably capture transient hemodynamic signatures in deep arteries. To address this challenge, we present a near-field resonant sensing system featuring a flexible, conformal interface for continuous arterial monitoring. The platform integrates a planar spiral single-coil sensor fabricated using flexible printed circuit (FPC) technology with a continuous-tracking readout architecture, enabling stable conformal coupling to the radial artery region and continuous acquisition of resonance-frequency shifts. Comparative experiments with a conventional rigid FR-4 interface demonstrated that the flexible FPC interface substantially improved coupling stability, reduced baseline drift induced by motion and deformation, and enhanced the beat-to-beat repeatability of cardiac-cycle waveforms. The resonance-derived signals showed strong agreement with synchronized PPG reference signals (Pearson’s r = 0.909, P < 0.001) while preserving subtle morphological features, including the dicrotic notch, that are essential for hemodynamic characterization and AI-based cardiovascular assessment. These findings identify conformal mechanical coupling as a key determinant of signal fidelity and robustness in near-field physiological sensing. Collectively, this work establishes a flexible physical front end for the continuous, non-invasive monitoring of arterial hemodynamics and provides a high-quality data foundation for AI-enabled cardiovascular screening and personalized intervention in precision medicine.

Keywords: cardiovascular diseases, arterial hemodynamics, near-field resonant sensing, continuous monitoring, flexible interface

Presentation number: PP 23

Abstract number: 2-ISABS-2026

ARTIFICIAL INTELLIGENCE–BASED FUNCTIONAL CLASSIFICATION OF PARA-SWIMMERS WITH ACHONDROPLASIA: A BIOMECHANICAL APPROACH FROM EULERIA LAB

Zabo Silva^1,1

¹Neuroxen – kinesiology and swimming, Poreč, Croatia

silvazabo@gmail.com

Artificial Intelligence–Based Functional Classification of Para-Swimmers with Achondroplasia: A Biomechanical Approach from EULERIA LAB. The goal of this study was to develop and validate an artificial intelligence–based model for objective functional classification of para-swimmers with achondroplasia using biomechanical and anthropometric parameters. A cohort of competitive para- swimmers with achondroplasia underwent three-dimensional kinematic analysis of stroke cycles, assessment of joint range of motion, limb proportions, and propulsion-related variables, combined with wearable sensor and video-based data acquisition in controlled swimming trials. Machine learning algorithms were trained to identify movement patterns and functional limitations relevant to swimming performance and to predict optimal sport class allocation. Model performance was evaluated against expert-based classification outcomes and competition results. The results demonstrated that the AI model could accurately discriminate functional profiles and showed high agreement with current classification decisions, while additionally revealing subtle biomechanical determinants of propulsion efficiency, stroke symmetry, and start and turn performance specific to the morphological characteristics of achondroplasia. Feature importance analysis indicated that upper-limb lever arm length, shoulder range of motion, and trunk stability were key predictors of functional capacity in water. In conclusion, the integration of artificial intelligence and biomechanical analysis provides an objective, reproducible, and transparent framework for functional classification of para-swimmers with achondroplasia, with potential to reduce subjectivity, improve fairness in competition, and support evidence-based refinement of classification systems, as well as individualized training and rehabilitation strategies developed within the EULERIA LAB.

Keywords: artificial intelligence, para-swimming, achondroplasia, functional classification, biomechanics

Presentation number: PP 24

Abstract number: 43-ISABS-2026

FROM PLANNING TO ROUTINE PRACTICE: DIGITAL TRANSFORMATION OF A PATHOLOGY DEPARTMENT AT UNIVERSITY HOSPITAL OF SPLIT

Zekić Tomaš Sandra^1,2, Kunac Nenad¹, Dunatov Huljev Ana¹, Anđelinović Šimun¹

¹Department of Pathology, Forensic Medicine and Cytology, University Hospital of Split, Split, Croatia; ²School of Medicine University of Split, Split, Croatia

szekic@mefst.hr

The digitalization of pathology departments represents a major step toward improving diagnostic efficiency, data accessibility, and the integration of advanced technologies in routine clinical practice. This study presents the implementation of a fully digital workflow at the Clinical Department of Pathology, University Hospital of Split. The digitalization process began in December 2023, following a structured planning phase that included defining the implementation timeline and technical requirements, and was completed in May 2025, after which the system has been continuously used in routine practice. The goal was to optimize workflow, reduce turnaround time (TAT), enable remote consultations, and establish a foundation for the application of AI tools. Materials and methods included the introduction of whole slide imaging (WSI), integration with PACS and LIS systems, and the reorganization of laboratory workflow starting from grossing and slide preparation, followed by prescan, selection of areas of interest, and scan quality control. This approach resulted in improved quality of hematoxylin and eosin (H&E) slides, which directly enhanced the quality of scanned WSI and contributed to overall laboratory quality improvement. The implementation required close collaboration between pathologists, the hospital IT department, the PACS provider, and scanner vendors, along with continuous staff training. Results demonstrated a significant improvement in workflow performance and TAT. The implementation also enabled improved collaboration, remote diagnostics, and increased satisfaction among staff. In conclusion, the digitalization of a pathology department is a complex, multidisciplinary process that extends beyond the mere acquisition of scanning equipment. It requires thorough assessment of technical infrastructure, coordinated involvement of multiple stakeholders, and continuous education to ensure successful implementation and long-term sustainability.

Keywords: digital pathology, whole slide imaging (WSI), workflow optimization, telepathology, artificial intelligence

Published: June 16^th, 2026.

Copyright: © 2026 Authors of Ai For Clinical Decision Support section. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.