Publications
Fine-Tuning SAM2 for Coronary Artery Segmentation in X-Ray Fluoroscopy
Sivakumar, E.
We fine-tune MedSAM2 on annotated coronary angiograms and apply it to video data for point-of-care use. On the ARCADE validation set (200 images), the fine-tuned model achieves Dice 0.767 ± 0.082 compared to 0.033 zero-shot. Applied across 10 fluoroscopic video studies, it tracks vessels coherently and avoids false segmenting of ribs, stents, and bypass grafts in 9 of 10 studies.
Analysis of Augmentation Techniques for Spine X-Ray Images
Sivakumar, E., Anjana Anand
We implement data augmentation strategies to tackle dataset imbalance in the VinDr-SpineXR dataset. Geometric transformations and GAN-based synthetic image generation are applied to abnormal classes, with classifier performance validated using VGG-16 and InceptionNet. A hybrid augmentation technique achieves ~99% validation accuracy with both classifiers across all three case studies.
Analysis of Machine Learning, Deep Learning, and Artificial Neural Network Approaches for Breast Cancer Classification
Sivakumar, E., Anjana Anand, Sachin Sarate
Breast cancer is one of the most common causes of death worldwide among women, with good survival rates if detected early. We compared supervised, semi-supervised and unsupervised learning on the Wisconsin Breast Cancer Dataset to establish the best-performing model for computer-aided diagnosis, with the goal of closing the gap between technology innovation and clinical implementation.
Episodic Recall in Autonomous Agents
Sivakumar, E., Kavitha Anand
Encoding and recall in autonomous agents have formed the crux of efficient design of intelligent systems. In this paper, we explore neuro-physiological modeling, reinforcement learning and Adaptive Resonance Theory implementation on mobile autonomous agents. We additionally test the performance of emotion-based encoding and retrieval of a 2D virtual input to an Adaptive Resonance Theory network.