Official implementation of "UniMedVL: Unifying Medical Multimodal Understanding and Generation through Observation-Knowledge-Analysis" - A unified medical vision-language model that integrates multimodal understanding and generation capabilities.

Automated segmentation, phenotyping, and spatial analysis of retinal flat-mount images with reproducible study workflows and publication-ready outputs.

AI for retinal health to classify AMD, assess fundus image quality, and generate synthetic retina images with PyTorch.

AI-powered diabetic retinopathy detection using ResNet-152 deep learning. Web app with Flask REST API, PyTorch, real-time retinal image classification, 5-stage severity detection. Medical imaging, computer vision, transfer learning, healthcare AI, ophthalmology screening tool.

Deep learning–based Glaucoma detection using CNN, Transfer Learning, and NSGA-II optimization, with a Flask web app for real-time predictions and Grad-CAM visualizations.

Web app for age-related macular degeneration (AMD) detection from retinal fundus images, with image quality assessment and Grad-CAM visual explanations.

Computational Stability of Cubical Homology: Orthogonal Controls for Fault Isolation in Retinal Diagnostics

takes Heidelberg DICOM export files and opens a heyex style viewer allowing for bscan to enface marking

Systematic safety audit of a ResNet-50 diabetic retinopathy classifier using the Medical Algorithmic Audit framework (Liu et al., 2022). Included error analysis, subgroup testing, adversarial robustness, and FMEA risk scoring on the APTOS 2019 dataset.

Detection and classification of diabetic retinopathy stages from retinal fundus images using GLCM texture feature extraction, Student's T-test feature selection, and SVM — 91.1% test accuracy, 95% AUC

Quantitative photometric and radiometric framework for simulating retinal photopigment bleaching and optoretinographic stimulus design.

DF41 team solution for the MICCAI 2024 MARIO Challenge. OCT-based AMD progression analysis using fusion CNNs and masked autoencoders. Ranked 2nd in Task 2 and 7th in Task 1.

Deep learning system for non-invasive cardiovascular risk prediction using retinal fundus images. Hybrid EfficientNet-B3 + ViT with clinical data fusion.

Deep learning–based system for automatic detection and severity classification of Diabetic Retinopathy from retinal fundus images using transfer learning and CNN ensembles.

AI/ML computational biomedical portfolio covering neurological signal analytics, retinal-image cardiovascular risk prediction, and DNA sequence-based gene expression modeling.