A novel diffusion-based method for weakly supervised semantic segmentation, achieving SOTA performance on fibrotic lung disease using only image-level labels and eliminating the need for pixel-level annotations.

Trained and Evaluated CNNs-based Segmentation Model for automated lung region segmentation.

Investigated the intermediate latent spaces in generative AI StyleGAN and Diffusion models for controlable image generation.

Evaluated cutting-edge GANs-based AI and their applications. Explored advancements in GAN architectures and their practical implementations across various domains

Developed a stereo vision-based 3D reconstruction pipeline for instrument tracking and measurement with 95% accuracy, incorporating feature detection, fluoroscopic tracking, and 3D volume rendering on surgical and zebrafish imaging data.

Retinal vessel segmentation using classical computer vision techniques—edge detection, image filtering, and ROC-based threshold selection.