| Scholar Name | Bhautik Daxini |
| Department | Instrumentation & Control |
| Reseach Area | Deep Learning |
| Supervisor Name | Dr. M.K.Shah |
| Title | Decision support system for lunds and Covid19 |
| Abstract | This thesis addresses the significant challenges in abdominal medical imaging by proposing an advanced deep learning framework for accurate and efficient multi-organ segmentation from CT scans. Medical imaging plays a crucial role in diagnostics, surgical planning, and disease monitoring. However, accurate segmentation of multiple organs from abdominal CT scans is complicated by high anatomical variability, overlapping tissue structures, subtle boundaries, and significant class imbalance among different organ tissues. Traditional segmentation techniques, while effective to some extent, often lack the necessary precision and computational efficiency to handle these complexities in a practical clinical setting. The proposed research introduces a novel hybrid approach integrating a modified cascaded V-Net architecture with Generative Adversarial Networks (GANs). This approach leverages the strengths of V-Net, renowned for handling volumetric medical data through its fully convolutional 3D structure, and GANs, recognized for their ability to generate realistic segmentation outputs and handle class imbalance by refining boundary delineations. By modifying the traditional V-Net to reduce computational demands and integrating a lightweight GAN discriminator, this research achieves significant improvements in segmentation accuracy, particularly around complex organ boundaries, while maintaining computational efficiency suitable for accessible GPU hardware. Extensive experimentation is conducted using well-established public datasets, specifically the BTCV (Beyond the Cranial Vault), LiTS (Liver Tumor Segmentation) and IRCAD (Liver segmentation 3D-IRCADb-01) databases. Comprehensive evaluations compare the proposed model against several state-of-the-art methodologies, including the 3D U-Net, ResU-Net, Swin Transformer, and UNETR. Quantitative analysis employs multiple metrics, with the Dice coefficient as the primary measure, alongside comprehensive resource utilization assessments such as GPU memory consumption, computat |