Medical image analysis
Deep learning for medical image analysis
Doctoral research at the University of Hyderabad on segmentation and classification of medical images — chiefly breast ultrasound — where the aim throughout was state-of-the-art accuracy at a parameter count that can actually be deployed in a clinic.
Research threads
4 areas
- 01 Advanced segmentation architectures EfficientU-Net and UMA-Net. Parameter-efficient encoders built from depthwise separable convolutions, atrous convolution blocks for adaptive receptive fields, and compound scaling across width, depth and resolution.
- 02 Loss function design The Fuzzy Rough Set (FRS) loss, and dynamically weight-adjusted ensemble losses for segmentation where boundaries are genuinely uncertain.
- 03 Multi-modal classification Saliency-guided attention networks for classification across medical imaging modalities.
- 04 Optimisation and feature selection Ensemble models combining deep features with genetic-algorithm feature selection, tuned for accuracy against parameter count and training time.
Publications
7 papers · newest first
- 01 Adaptive ensemble loss and multi-scale attention in breast ultrasound segmentation with UMA-Net Dar, M.F. & Ganivada, A. DOI
- 02 Deep learning and genetic algorithm-based ensemble model for feature selection and classification of breast ultrasound images Dar, M.F. & Ganivada, A. DOI
- 03 EfficientU-Net: A novel deep learning method for breast tumor segmentation and classification in ultrasound images Dar, M.F. & Ganivada, A. DOI
- 04 Dynamic weight-adjusted ensemble loss for enhanced medical image segmentation Dar, M.F. & Ganivada, A. DOI
- 05 Design and analysis of a robust security layer for software defined network framework Alhaj, A.N., Patel, N.D., Singh, A., Bondugula, R.K., Dar, M.F. & Ahamed, J. DOI
- 06 Latent fingerprint enhancement and matching using intuitionistic type-2 fuzzy Mukhtar, S., Dar, M.F. & Kaur, A. DOI
- 07 Performance comparison of face detection and recognition algorithms Dar, M.F. & Dixit, S. DOI
Every entry links to its DOI. The Google Scholar profile above is the authoritative, current list.
Presentations
The PhD research presentation is not currently published on this portal.
The papers above cover the same work, and each links to its published version.