Publications

For a full list, see my Google Scholar profile.

Selected Publications

2026

Wave-GMS: Lightweight Multi-Scale Generative Model for Medical Image Segmentation

ICASSP 2026

Talha Ahmed, Nehal Ahmed Shaikh, Hassan Mohy-ud-Din

A multi-resolution generative model for medical image segmentation with only ~2.6M parameters, achieving state-of-the-art performance and strong cross-domain generalization. We combine a Haar wavelet-based encoder (ℰ_wave) with a compact mask tokenizer (ℰ_tiny) to eliminate reliance on large pretrained vision foundation models such as SD-VAE. A latent-alignment loss (ℒ_align) ensures that LMM produces codes consistent with the decoding space of 𝒟_tiny.

arXiv GitHub

In Preparation

Unified Perspective on Diffusion Models: Theory and Practice in Medical Imaging and Inverse Problems

Survey (In Preparation)

Talha Ahmed, Nehal Ahmed Shaikh

A comprehensive survey unifying theoretical perspectives on diffusion models (autoencoders, flow, energy, score-based methods) and their applications in imaging and inverse problems. Analyzes methods for segmentation, restoration, super-resolution, and generation (e.g., ControlNet, SDSeg, StableSR, ADD), highlighting fine-tuning, cross-modality learning, and latent space optimization strategies.

Draft