Hi, I'mAbhay Prabhakar

Brain tumor segmentation is a challenging and critical task in the area of medical imaging and treatment planning. The result of the segmentation process is to generate an accurate sketch of the tumor regions by their corresponding brain tumor masks. The precise and accurate segmentation of tumors requires multiple image modalities with varying contrasts and spatial resolution. In recent years, many deep learning models such as U-Net, ResNet, VGG16, and Transformer-based architectures have emerged and achieved promising results with exceptional breakthroughs. The survey has been utilized to review state-of-the-art deep learning methods for brain tumor segmentation, clearly focusing...

Co-Authors

Dr. Archana BhatAadishesh Gopal PadasalgiAditya B

Brain Tumor SegmentationDeep LearningMedical ImagingU-NetComputer VisionAI in Healthcare

Diabetic retinopathy is one of the leading causes of preventable blindness worldwide. However, screening for this condition is traditionally performed through manual examination of fundus images, which presents several challenges. This paper describes an automated diabetic retinopathy severity grading system based on EfficientNet-B0, a lightweight convolutional architecture with only 4.3 million parameters that offers a balance between computational efficiency and diagnostic accuracy for five-class DR grading. The proposed framework applies CLAHE enhancement in the LAB color space during preprocessing, which improves the visibility of retinal features while preserving color information. Training is performed using WeightedRandomSampler to address class imbalance, along with label-smoothing cross-entropy loss and AdamW optimization. The model is trained and evaluated using the combined Messidor-2 and APTOS 2019 datasets. The system achieves an accuracy of 84.95% on the held-out test set, with a quadratic weighted Cohen's Kappa of 0.9103, reflecting strong agreement with expert grading. Per-class performance is particularly strong for healthy retinas, achieving 93.48% accuracy, and for moderate DR cases, achieving 81.51%. Grad-CAM visualizations further confirm that the model's predictions are guided by clinically relevant features, such as hemorrhages and exudates, rather than spurious correlations. While the current evaluation is limited to internal test splits and further external validation across diverse populations is warranted, this lightweight end-to-end pipeline demonstrates strong potential as a clinical decision-support tool to assist ophthalmologists in patient triage.

Co-Authors

Dr. Archana BhatAdithya SAditya BAadishesh Gopal Padasalgi

Diabetic RetinopathyEfficientNet-B0Grad-CAMComputer VisionAI in HealthcareMedical ImagingMachine Learning