About

Amal's research focuses on the applications of artificial intelligence to improve healthcare, particularly in cardiovascular disease care. He has expertise in deep learning, machine learning, natural language processing (NLP), image processing, and recommender systems. His work aims to enhance healthcare outcomes through innovative technological solutions.

Research topics

• Artificial Intelligence
• Computer Science
• Machine Learning
• Medicine
• Data Mining
• Data science
• Human–computer interaction
• Database
• Bioinformatics
• Internal medicine
• Medical physics
• Family medicine
• Radiology

Selected publications

• Genome-Wide Association Study of Genetic Variants Associated with Lower Extremity Amputation Risk in Peripheral Artery Disease

  Preprints.org · 2026-02-11

  preprintOpen accessSenior author

  Peripheral artery disease (PAD) is a global health burden affecting over 200 million individuals and is frequently complicated by limb-threatening ischemia, leading to major amputations. Despite known clinical risk factors, the genetic basis underlying amputation risk in PAD remains poorly defined. In this study, we performed a multi-pronged genome-wide association study (GWAS) to identify genetic variants associated with lower extremity amputation in patients with PAD, using data from the All of Us Research Program. Two analytical strategies were employed: a targeted GWAS using ClinVar variants on the full cohort and a comprehensive genome-wide association study using Allele Count/Allele Frequency (ACAF) data on a balanced subset. The ClinVar analysis of 118,871 variants in 14,771 PAD patients (613 with amputation, 14,158 without) identified 3 suggestive associations with a genomic inflation factor of 1.046. The ACAF analysis of 7,784,837 quality-controlled variants in 804 balanced samples (399 cases, 405 controls) yielded 35 suggestive associations (p < 1×10⁻⁵) with a genomic inflation factor of 1.017. No variants achieved suggestive significance in both analyses. These findings highlight candidate loci for further validation and may inform future development of risk prediction tools and targeted interventions to reduce limb loss in PAD.

  PublisherOA PDFDOI

• Vision Transformers Based AI Models For Predicting Colorectal Cancer from Digital Pathology WSI: Use Case Of MHIST dataset

  medRxiv · 2026-02-04

  articleOpen accessSenior authorCorresponding

  Abstract This study investigates the efficacy of transformer-based deep learning architectures—specifically, Vision Transformer (ViT), Class Attention in Image Transformers (CaiT), and Data-Efficient Image Transformers (DeiT)—for the binary classification of colorectal polyps using the Minimalist Histopathology Image Analysis Dataset (MHIST). The dataset comprises 3,152 hematoxylin and eosin (H&E)-stained Formalin Fixed Paraffin-Embedded (FFPE) images annotated as either Hyperplastic Polyps (HP) or Sessile Serrated Adenomas (SSA). A rigorous evaluation was conducted using a 5-fold stratified cross-validation methodology, and performance was quantified using metrics including accuracy, precision, recall, F1-score, and AUC-ROC. Experimental results revealed that transformer architectures, particularly CaiT (accuracy of 90.18%, AUC-ROC of 95.52%), outperformed traditional convolutional neural networks (CNNs). The superior performance of CaiT is attributed to its specialized class-attention mechanisms, effectively capturing nuanced morphological differences essential for accurate histopathological classification. These findings underscore the potential of transformer-based models to enhance diagnostic precision, reduce variability in pathological assessment, and facilitate earlier and more reliable colorectal cancer screening.

  PublisherOA PDFDOI

• Confounder Analysis and Visualization of Treatment Pathways Associated with Amputation in Patients with Diabetes Using Sankey Diagrams: Enhancing Explainability

  SSRN Electronic Journal · 2026-01-01

  preprintOpen accessSenior author
  PublisherDOI

• An Agentic System for Natural Language Querying of The Cancer Genome Atlas Clinical Data

  Research Square · 2025-10-10

  preprintOpen accessSenior author
  PublisherOA PDFDOI

• Digital Pathology and Ensemble Deep Learning for Kidney Cancer Diagnosis: Dartmouth Kidney Cancer Histology Dataset

  Applied Biosciences · 2025-02-05 · 6 citations

  articleOpen accessSenior author

  Kidney cancer has become a major global health issue over time, showing how early detection can play a very important role in mediating the disease. Traditional histological image analysis is recognized as the clinical gold standard for diagnosis, although it is highly manual and labor-intensive. Due to this issue, many are interested in computer-aided diagnostic technologies to assist pathologists in their diagnostics. Specifically, deep learning (DL) has become a viable remedy in this field. Nonetheless, the capacity of existing DL models to extract comprehensive visual features for accurate classification is limited. Toward the end, this study proposes using ensemble models that combine the strengths of multiple transformers and deep learning model architectures. By leveraging the collective knowledge of these models, the ensemble enhances classification performance and enables more precise and effective kidney cancer detection. This study compares the performance of these suggested models to previous studies, all of which used the publicly accessible Dartmouth Kidney Cancer Histology Dataset. This study showed that the Vision Transformers, with an average accuracy of over 99%, were able to achieve high detection accuracy across all complete slide picture patches. In particular, the CAiT, DeiT, ViT, and Swin models outperformed ResNet. All things considered, the Vision Transformers consistently produced an average accuracy of 98.51% across all five-folds. These results demonstrated that Vision Transformers might perform well and successfully identify important features from smaller patches. Through utilizing histopathological images, our findings will assist pathologists in diagnosing kidney cancer, resulting in early detection and increased patient survival rates.

  PublisherOA PDFDOI

• MedSAM and Generative AI for segmenting gleason pattern in digital pathology slides of prostate cancer

  2025-06-20

  preprintOpen accessSenior author

  Prostate cancer is a leading cause of cancer-related mortality among men, and precise grading of histopathology slides is critical for treatment planning. We introduce a prompt-guided adaptation of the Segment Anything Model (MedSAM) for pixel-level, multi-class Gleason pattern segmentation in haematoxylin–eosin-stained tissue micro-arrays. Using stratified train/validation/test splits of the MICCAI-2019 dataset, the network combines class-aware oversampling, colour-preserving augmentation and a weighted cross-entropy + Dice objective to counter severe class imbalance. On the held-out test set our model achieves Cohen’s κ =0 . 772, F macro = 0 . 8 1 3 and F micro = 0 . 8 5 7 . Averaged over five Gleason grades, sensitivity and specificity reach 0.882 and 0.953, respectively, surpassing strong CNN (U-Net, DeepLab) and transformer baselines (K-Net, Swin-UNet). In particular, high-grade patterns (grades 4–5) are recalled with 0.909 sensitivity while benign tissue is segmented with > 90 % specificity. The system processes a 1024×1024 patch in 128 ms on a single V100 GPU, underscoring the practicality of fine-tuned foundation models as fast and reliable decision-support tools in digital pathology workflows. Our objective is to build a clinically usable, multi-grade Gleason segmentation tool by fine-tuning MedSAM with prompt guidance and imbalance-aware learning.

  PublisherOA PDFDOI

• Artificial Intelligence and Digital Pathology for Histologic Growth Pattern Classification in Lung Adenocarcinoma

  SSRN Electronic Journal · 2025-01-01

  preprintOpen accessSenior author
  PublisherDOI

• Deep Learning Models for Automated Classification of Seborrheic Keratosis: A Comprehensive Literature Review and Comparative Study

  Preprints.org · 2025-05-23

  preprintOpen accessSenior author

  Seborrheic keratosis (SK) is a common benign skin lesion that is often clinically mistaken for malignant melanoma due to visual similarities. This misdiagnosis can lead to unnecessary patient anxiety, invasive procedures, and increased healthcare costs. Deep learning models have recently shown promise in improving the accuracy and objectivity of skin lesion diagnosis. In this work, we present a comprehensive literature review of automated SK classification using deep learning, and we perform a comparative study of four state-of-the-art architectures: ResNet-34, EfficientNet-B1, Vision Transformer (ViT), and VGG16, on multi-source dermoscopic image datasets. Our literature review highlights that, while most prior studies focused on melanoma detection or general skin lesion classification, relatively few have specifically addressed SK, which is a significant gap given its prevalence and propensity for misdiagnosis. We summarize key contributions from recent research, including convolutional neural network (CNN) approaches and emerging transformer-based models for skin lesion analysis. In our experimental evaluation across three diverse datasets (DermoFit, BCN20000, and an Argentine clinical dataset), we found that individual model performance varies widely. ResNet-34 achieved a high area under the ROC curve (AUC) of 0.9742 with strong specificity, and EfficientNet-B1 attained the highest validation accuracy (94.41%) among the CNNs. A Vision Transformer model, after careful tuning and augmentation, outperformed the CNNs, achieving a test accuracy of 97.28% on the SK classification task. This improved ViT model demonstrated a balanced sensitivity and specificity (both above 95%), underscoring the potential of transformer architectures in skin lesion classification. We discuss these results in the context of existing literature and clinical requirements. Overall, our study provides an up-to-date review of deep learning techniques for SK identification and emphasizes the value of transformer-based models for this challenging dermatological problem.

  PublisherOA PDFDOI

• Ensemble Deep Learning Models for Early Detection of Meningitis in ICU: Multi-center Study

  ArXiv.org · 2025-10-17

  preprintOpen accessSenior author

  The stacking ensemble combining RF, LightGBM, and DNN performed well on internal test sets, exhibiting an NPV greater than 99.9% even with substantial class imbalance. While performance was lower on the external eICU cohort compared to the internal test sets, sensitivity remained robust. Therefore, the stacking ensemble may serve as a rule-out screening option for ERs and ICUs after additional prospective multi-site validation studies for its efficacy in real-world.

  PublisherOA PDFDOI

• Genome-Wide Association Study of Genetic Variants Associated with Lower Extremity Amputation Risk in Peripheral Artery Disease

  International Journal of Molecular Sciences · 2025-12-20

  preprintOpen accessSenior authorCorresponding

  Abstract Peripheral artery disease (PAD) is a global health burden affecting over 200 million individuals and is frequently complicated by limb-threatening ischemia, leading to major amputations. Despite known clinical risk factors, the genetic basis underlying amputation risk in PAD remains poorly defined. In this study, we performed a multi-pronged genome-wide association study (GWAS) to identify genetic variants associated with lower extremity amputation in patients with PAD, using data from the All of Us Research Program. Two analytical strategies were employed: a targeted GWAS using ClinVar variants on the full cohort and a comprehensive genome-wide association study using Allele Count/Allele Frequency (ACAF) data on a balanced subset. The ClinVar analysis of 118,871 variants in 14,771 PAD patients (613 with amputation, 14,158 without) identified 3 suggestive associations with a genomic inflation factor of 1.046. The ACAF analysis of 7,784,837 quality-controlled variants in 804 balanced samples (399 cases, 405 controls) yielded 35 suggestive associations (p < 1×10□□) with a genomic inflation factor of 1.017. No variants achieved suggestive significance in both analyses. These findings highlight candidate loci for further validation and may inform future development of risk prediction tools and targeted interventions to reduce limb loss in PAD.

  PublisherOA PDFDOI

Frequent coauthors

• Anne Breggia

  12 shared

• Salah Alheejawi

  Northeastern University

  10 shared

• Robert Christman

  Maine Medical Center

  10 shared

• Bilal Ahmad

  Maine Medical Center

  9 shared

• Elsie Gyang Ross

  7 shared

• Einat Minkov

  7 shared

• Akarsh Singh

  Northeastern University

  6 shared

• Tsvi Kuflik

  University of Haifa

  6 shared

Education

• B.S., Computer Science

  Technion

  2009

• M.S., Computer Science

  University of Haifa

• Ph.D., Computer Science

  University of Haifa