Medical Image Analysis · Ultrasound · Uncertainty-Aware Deep Learning
I am an Assistant Professor in the School of Computer Science at UPES Dehradun. I build deep learning for ultrasound images that treats ambiguity as a signal rather than noise — fuzzy rough set losses for uncertain lesion boundaries, and lightweight, interpretable models that stay reliable on small, heterogeneous datasets.
Citations per year
2023: 3 citations, 2024: 16 citations, 2025: 48 citations, 2026: 47 citations
Source: Google Scholar, updated weekly.
MSCT-Trans — a multi-scale CNN token transformer for interpretable breast, thyroid and fetal ultrasound classification — published online in Ultrasound in Medicine & Biology (vol. 52, no. 11). Read the paper
Saliency-guided AttentionNet, a dual-branch model for breast ultrasound classification, published online in Biomedical Signal Processing and Control (vol. 127). Read the paper
Awarded the PhD degree by the University of Hyderabad on 28 July 2026, for the thesis “Advances in Deep Learning for Medical Image Segmentation and Classification”. Read the summary
Moved to the Research Faculty track in the School of Computer Science, UPES.
“Fuzzy Rough Set Loss for Deep Learning-Based Precise Medical Image Segmentation” published online in Computerized Medical Imaging and Graphics (vol. 128). Read the paper
Joined the School of Computer Science, UPES Dehradun, as Assistant Professor.
Submitted my PhD thesis, “Advances in Deep Learning for Medical Image Segmentation and Classification”, to the University of Hyderabad.
UMA-Net — adaptive ensemble loss and multi-scale attention for breast ultrasound segmentation — published online in Medical & Biological Engineering & Computing. Read the paper
Presented “Dynamic Weight-Adjusted Ensemble Loss for Enhanced Medical Image Segmentation” at ICCCNet-2024 in Manchester, UK, supported by a University of Hyderabad IoE International Travel Grant.
Our work on genetic algorithm-based feature selection for breast ultrasound classification was published in Image and Vision Computing. Read the paper
Ultrasound in Medicine & Biology, 52(11), 2623–2638, 2026
Multi-scale convolutional features from a lightweight backbone become a unified token sequence for a transformer encoder, giving cross-scale global reasoning and interpretable predictions across breast, thyroid and fetal ultrasound.
Biomedical Signal Processing and Control, 127, 111194, 2026
Grad-CAM saliency splits each scan into lesion and context branches over a shared backbone, fused by an adaptive attention block — 90.51% accuracy across five public datasets and 78.46% on held-out data.
Computerized Medical Imaging and Graphics, 128, 102716, 2026
A loss built on fuzzy rough set theory that treats boundary ambiguity as a first-class signal, combining fuzzy similarity with lower and upper approximations — a 2.1% average Dice gain over the best baseline across breast ultrasound, polyp, brain MRI, chest CT and skin-lesion datasets.
Medical & Biological Engineering & Computing, 63(6), 1697–1713, 2025
Residual connections, attention blocks and atrous convolutions trained under a dynamic ensemble loss that rebalances BCE, Dice, Hausdorff and Tversky terms during training — generalising across five breast ultrasound datasets.
Image and Vision Computing, 146, 105018, 2024
Deep features selected by a genetic algorithm and classified by a weighted-voting ensemble, gaining 4–9% accuracy on benchmark breast ultrasound data while curbing overfitting on small datasets.
Neural Processing Letters, 55(8), 10439–10462, 2023
EfficientNet-B7 and atrous convolution inside U-Net: a 13× parameter reduction over U-Net (1.31M vs 17.27M) with 97.9% classification accuracy on breast ultrasound.
01
Loss functions that treat boundary ambiguity as signal rather than noise. The fuzzy rough set loss combines fuzzy similarity with lower and upper approximations, improving Dice by 2.1% on average over the best baseline across breast ultrasound, polyp, brain MRI, chest CT and skin-lesion datasets (CMIG, 2026). An adaptive ensemble loss rebalances BCE, Dice, Hausdorff and Tversky terms during training (MBEC, 2025).
02
Lightweight models that show where they look. Saliency-guided AttentionNet models lesion and surrounding tissue in separate branches: 90.51% accuracy across five public breast ultrasound datasets, and 78.46% on held-out data (BSPC, 2026). MSCT-Trans turns multi-scale CNN features into transformer tokens for breast, thyroid and fetal ultrasound, with Grad-CAM++ localisation and confidence analysis (UMB, 2026).
03
Current work on pixel-level uncertainty estimation, aimed at flagging unreliable segmentations so that only uncertain cases go to human review.