Dissertations - M Tech (CS)
Permanent URI for this collectionhttp://164.52.219.250:4000/handle/10263/2147
These Dissertations were submitted in partial fulfilment of the requirements for the award of M TECH (Computer Science) Degree of Indian Statistical Institute
Browse
6 results
Search Results
Item A Novel Approach to Medical Image Segmentation with Convformer-Based Attention Mechanism and UNet(Indian Statistical Institute, Kolkata, 2024-06) Nandi, SwastikAccurate segmentation of medical images is a critical task in the field of healthcare, aiding in precise diagnosis and effective treatment planning. This project explores the enhancement of image segmentation models through the integration of advanced attention mechanisms. Our primary objective is to compare various attention techniques to develop a lightweight yet highly accurate model suitable for real-time applications. Given the significant body of work in medical image segmentation, our approach seeks to balance accuracy with computational efficiency. By incorporating different attention mechanisms and rigorously evaluating their performance, we aim to identify the optimal strategy for improving segmentation outcomes. The results demonstrate the potential for improved segmentation accuracy and efficiency, highlighting the effectiveness of attention-based models in capturing intricate patterns and dependencies within medical imaging data. We found out in our work that the CNN-based attention mechanism, or Convformer, effectively overcomes the issues related to the training conflict between CNNs and transformers. This project sets the groundwork for future advancements in semi-supervised and weakly-supervised learning, and we plan to expand our model’s applicability across a broader range of medical imaging scenarios. Our ultimate objective is to contribute towards the development of robust, efficient, and adaptable segmentation models that can enhance diagnostic accuracy and patient care in various medical fields.Item Few shot segmentation for COVID-19 infected lung CT slices(Indian Statistical Institute, Kolkata, 2022-07) Chatterjee, SohamThe last 2 years have been adversely affected by the COVID-19 pandemic. Doctors usually detect Covid from CT slices from features such as ground glass, consolidation and pleural effusion. These features usually have complex contours, irregular shapes and rough boundaries. With increasing number of cases the workload on the radiologists have increased by leaps and bounds to analyze the lung CT scans for tracking the disease progression in the patient. Moreover manual analysis of the CT scans is also prone to human error. So automated segmentation of infected lung CT slices can help the doctors to diagnose the disease faster. With the advent of deep learning, various approaches have been built to tackle this problem of automated biomedical image segmentation. One such architecture is the U-Net by Ronnenberger et al. [14]. Various other approaches have been proposed which are all variations of the U-Net to achieve better segmentation performance. However, the U-Net and its variations suffer from high model complexity, due to which they easily overfit on limited labelled dataset which is a serious issue in medical image domain. To cater this problem of data scarcity, research in “few shot segmentation” has gained significant importance in the recent years. In this work, we have developed a deep neural network model called Few Shot Conditioner Segmenter Covid (FSCS-cov), an architecture to tackle the problem of segmenting different COVID- 19 lesions from limited number of COVID-19 infected lung CT slices using few - shot learning paradigm.Item PIXEL CLASSIFICATION USING U-NET(Indian Statistical Institute, Kolkata, 2022) Chouhan, AyushThe rapid advances in Deep Learning (DL) techniques have allowed rapid detection, localisation, and recognition of objects from images or videos. DL techniques are now being used in many different applications related to agriculture and farming and medical Science Images. In this work we are using Deep Learning techniques such as unet,pretrained unet and apply on CWIF data set for Anomaly Detection and anomaly is weed and on Electron Microscopy Dataset we are detecting mitochondria in hippocampus region of the brain we evaluate our model using different losses and evaluation metrics at the same time also telling the drawback and advantages of different models. If we can detect the images in the crops we can use different machines that can be used for real time detection and removal of weed from the field Our technology can distinguish between crop and weed plants in commercial fields where crop and weed grow near to one another and can tolerate plant overlap. Automated crop/weed discrimination allows for targeted weed treatment in weed management tactics to reduce expense and adverse environmental effects. The images of hippocampus region of the brain to detect mitochondria in the images and give lable to each pixel will it belong to mitochondria or notItem Deep learning for COVID-19 lung pathology segmentation(Indian Statistical Institute, Kolkata., 2021-07) Bedi, Gurdit SinghCOVID-19 pandemic has impacted billions of lives and created a challenge for the healthcare systems. Detection of pathologies from computed tomography (CT) images offers a great way to assist the traditional healthcare for tackling COVID-19. Pathologies such as ground-glass opacification and consolidations are region of interests which the doctors use to diagnosis the patients. In this work, we have developed and tested various segmentation model using transfer learning to find such pathologies. U-Net [15] is the foundation of the models which we have tested. Along with U-Net we have changed the encoder section of the said model, to various classification models such as VGG, ResNet and MobileNet. As these model have won ImageNet Challenge, there core component have been used for feature extraction and usage of their pretrained weights will help in faster convergence. A small subset of studies which has been annotated with binary pixel masks depicting regions of interests in MosMedData [12] Chest CT Scans dataset have been used to train the segmentation model. The best segmentation model achieved a mean dice score of 0.6029.Item CNN for brain tumor segmentation(Indian Statistical Institute, Kolkata, 2017) Singhaniya, MohitItem CNN for brain tumor segmentation(Indian Statistical Institute, Kolkata, 2017) Singhaniya, Mohit