Moses Schanfield Young Investigator Award (MSYIA) Presentations
Presentation number: MSYIA 01
Abstract number: 67-ISABS-2026
DECODING TUMOR ORIGIN USING ARTIFICIAL INTELLIGENCE: TRANSFORMING DIAGNOSTIC STRATEGIES IN PRECISION ONCOLOGY
Brlek Petar1,2,3,4,5, Bulić Luka1,2,4,6, Shah Parth5, Primorac Dragan1,2,4,7,8,9,10,11,12,13,14
1St. Catherine Specialty Hospital, Zagreb, Croatia , Zagreb, Croatia; 2International Center for Applied Biological Research, Zagreb, Croatia , Zagreb, Croatia; 3Department of Molecular Biology, Faculty of Science, University of Zagreb, Zagreb, Croatia , Zagreb, Croatia; 4School of Medicine, Josip Juraj Strossmayer University of Osijek, Osijek, Croatia, Osijek, Croatia; 5Dartmouth Health, Lebanon, NH, USA , Lebanon, NH, United States of America; 6Algebra Bernays University, Zagreb, Croatia, Zagreb, Croatia; 7Eberly College of Science, The Pennsylvania State University, State College, PA, USA, State College, United States of America; 8School of Medicine, University of Split, Split, Croatia, Split, Croatia; 9The Henry C. Lee College of Criminal Justice and Forensic Sciences, University of New Haven, New Haven, CT, USA, New Haven, United States of America; 10Sana Kliniken Oberfranken, Coburg, Germany, Coburg, Germany; 11School of Medicine, University of Rijeka, Rijeka, Croatia, Rijeka, Croatia; 12Faculty of Dental Medicine and Health, Josip Juraj Strossmayer University of Osijek, Osijek, Croatia, Osijek, Croatia; 13School of Medicine, University of Mostar, Mostar, Bosnia and Herzegovina, Mostar, Bosnia and Herzegovina; 14National Forensic Sciences University, Gandhinagar, India, Gandhinagar, India
petar.brlek@svkatarina.hr; luka.bulic@svkatarina.hr
Cancers of unknown primary (CUP) remain a major clinical challenge due to limited diagnostic accuracy and the inability to apply optimal site-specific therapies, as most oncological treatment protocols are defined according to the primary tumor origin. In the era of precision oncology, characterized by the increasing use of whole-genome sequencing and artificial intelligence, we aimed to develop and validate an AI-driven approach for predicting the primary tumor site based on comprehensive genomic profiling, with the objective of improving diagnostic accuracy and supporting clinical decision-making. An in silico diagnostic study was conducted using publicly available datasets comprising over 20,000 metastatic tumor samples with annotated clinical variables and mutational profiles across more than 600 cancer-related genes. Multiple machine learning algorithms were trained and evaluated for multi-class tumor site classification, with model performance assessed using cross-validation and an independent test cohort. The optimal model was implemented into a software platform with an integrated graphical user interface (GUI) to facilitate clinical application. Among the evaluated models, the XGBoost-based classifier demonstrated superior performance, achieving a top-2 accuracy of 0.91 and an ROC-AUC of 0.97, with consistent generalizability across datasets. Feature importance analysis identified biologically relevant genomic patterns contributing to tumor classification. This approach enables more accurate tumor origin prediction and supports more informed, site-directed therapeutic decision-making, particularly in diagnostically challenging cases such as CUP. The developed platform represents a step toward translating complex genomic data into clinically actionable insights, while future integration with deep whole-genome sequencing and liquid biopsy approaches may further improve tumor origin identification and advance precision oncology.
Keywords: artificial intelligence, cancer of unknown primary, machine learning, next-generation sequencing, precision oncology
Presentation number: MSYIA 02
Abstract number: 82-ISABS-2026
TRANSFORMING CANCER DIAGNOSIS: QUANTUM AI FOR GLOBAL LEUKEMIA SCREENING
Rahman Ahona1, Saha Joty1, Hasan Babu Hafiz Mohammad2, Gupta Rajat Das
1Department of Computer Science and Engineering, University of Dhaka, Dhaka, Bangladesh; 2Vanderbilt University Medical Center, Nashville, TN, USA
hafizbabu@du.ac.bd
Acute lymphoblastic leukemia (ALL) is the most prevalent childhood malignancy, yet survival rates diverge sharply between high-income (>90%) and low-income settings (<50%), largely due to delayed diagnosis. Flow cytometry and cytogenetic analysis remain inaccessible across 70% of facilities in low – and middle-income countries. Classical AI approaches for automated blood smear analysis achieve limited accuracy (75-82%) with classical ML and require substantial computational resources for deep learning alternatives. We present QUANT-HemoDx, a hybrid quantum-classical framework integrating ResNet-18 convolutional feature extraction with an 8-qubit variational quantum circuit (VQC). Classical features (512-dimensional) are compressed and encoded into quantum states via parameterized RY rotations. Quantum superposition enables parallel morphological feature evaluation, entanglement captures higher-order cellular correlations, and interference reinforces diagnostically meaningful patterns. The model was trained on the C-NMC 2019 dataset (15,114 images, 118 patients) and independently validated at Ahsania Mission Cancer & General Hospital, Dhaka, Bangladesh. Hardware validation was performed on IBM’s ibm_kyoto 127-qubit quantum processor. QUANT-HemoDx achieved 96.7% accuracy (ROC-AUC 0.991; 95% CI: 0.981-0.998), sensitivity 98.1%, and specificity 95.2% on the benchmark test set. Clinical validation yielded 94.2% accuracy on real quantum hardware. Compared to ResNet-50, the architecture requires 266× fewer parameters (96,000 vs. 25.6 million) and trains 1.68× faster, reducing per-patient diagnostic cost to $0.08. QUANT-HemoDx demonstrates that quantum AI architectures can achieve clinically competitive ALL detection with dramatically reduced computational overhead – a meaningful advantage for resource-limited screening contexts. Prospective multicenter validation is ongoing.
Keywords: quantum AI, leukemia detection, cancer prevention, early diagnosis, health equity
Presentation number: MSYIA 03
Abstract number: 27-ISABS-2026
FROM SPEED TO TRUST: THE GOVERNANCE OF ARTIFICIAL INTELLIGENCE‑ASSISTED KNOWLEDGE PRODUCTION IN PRECISION MEDICINE
Todorović Petar1,2, Alfirević Nikša3, Maglica Mirko2, Vukojević Katarina1,2
1Department of Anatomy, Histology and Embryology, University of Split School of Medicine, Split, Croatia; 2Department of Anatomy, School of Medicine, University of Mostar, Mostar, Bosnia and Herzegovina; 3Faculty of Economics, Business and Tourism, University of Split, Split, Croatia
petar.todorovic@mefst.hr
Artificial Intelligence (AI) transforms knowledge-intensive work by generating outputs faster than they can be verified. The goal of this study was to examine the consequences of this asymmetry in science production, with a focus on its implications for precision medicine and biomedical research. In this paper, we analyze current evaluation systems reward speed instead of validation and deliver outputs at a pace that exceeds verification. Our analysis identifies structural deficiencies in institutions that may lead unverified AI-assisted claims to enter translational pipelines and clinical practice. In precision medicine, the risk of unverified AI claims increases, as AI-assisted research processes tend to produce statistically probable, consensus-aligned outputs that can suppress the biological heterogeneity precision medicine is designed to address. We propose a system that checks AI-assisted claims, which includes built-in staged review points and separation of teams that generate theories, and conduct empirical research, from those that evaluate their results. This challenge extends beyond technical issues. It requires organizational solutions and the formulation of an early model for structuring knowledge work in the AI era.
Keywords: AI governance, evaluation infrastructure, precision medicine, organizational design, knowledge production
Published: June 16th, 2026.
Copyright: © 2026 MSYIA authors. 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.
