SK Rakib Ul Islam Rahat

Graduate Researcher in Multimodal AI

I work on medical artificial intelligence with a focus on multimodal deep learning, medical image analysis, and probabilistic calibration. My research emphasizes conservative evaluation, robustness under dataset shift, and protocol-faithful reporting for clinical screening models.

Email: skrakibulislamrahat@gmail.com
Phone: +1 (323) 471-5980
Google Scholar: scholar.google.com/citations?user=0X1eRi8AAAAJ

SK Rakib Ul Islam Rahat

About

My work lies at the intersection of machine learning and healthcare. I focus on identifying and quantifying failure modes in medical AI, including miscalibration, shortcut learning, and performance degradation under external validation.

Research Interests

Education

Research Projects

Calibration Analysis for Referable Diabetic Retinopathy (IJBHI)

Multi-seed internal calibration analysis of a ResNet-18 fundus classifier trained on APTOS 2019 using Counterfactual Consistency Regularization, with constrained external evaluation on Messidor-2.

GitHub

Exposing Dataset Artifacts in Medical AI

Empirical study demonstrating shortcut learning driven by non-pathological artifacts in fundus imaging models.

GitHub

Lightweight DR Detection Models

Evaluation of MobileNetV2, EfficientNet-B0, and SqueezeNet focusing on accuracy–efficiency trade-offs for low-resource screening.

GitHub

Multimodal Framework Research

Multimodal pipelines integrating medical images and structured clinical data to study robustness and interpretability.

GitHub

Selected Publications

  1. Calibration Analysis Under Counterfactual Consistency Regularization for Referable Diabetic Retinopathy. IEEE JBHI (under review)
  2. TriGWONet: A Lightweight Multibranch CNN Using Gray Wolf Optimization. Discover AI, 2025
  3. Artificial Intelligence for Chronic Kidney Disease Risk Stratification. BJNS, 2025
  4. Advancing Diabetic Retinopathy Detection with AI. BJNS, 2025
  5. Deep Learning Framework for Early Breast Cancer Detection. BJNS, 2025