Enhancing Tumor Segmentation in 3D MRI Scans with YOLOv8 And Volumetric segmentation: A Computational Efficient Approach

Abstract

The computational aspect of this field also faces challenges. The high resolution and large data volumes of images require resources for processing. This does not require infrastructure but can also lead to delays in diagnosis and treatment planning, which is not ideal for medical professionals. Additionally, the costs associated with maintaining and acquiring systems pose a dilemma for healthcare organizations and research institutions. Even when computational challenges are addressed network designs like CNNs have their limitations. They may struggle to capture all the details in an image and techniques, like pooling can result in the loss of data. This research aims to overcome these challenges by focusing on optimizing the analysis of brain tumor images. We place importance on reducing the burden while still capturing crucial tumor specific information. To achieve this we introduced a pipeline that ensures reproducible models by standardizing the data. Our methodology is based on combining data processing, computer vision and learning techniques. We have innovatively integrated the YOLOv8 model to enable tumor localization and prediction in unexamined imaging datasets. Through the technique of data stacking we create three representations of MRI scans along with their masks. Training a model on these stacked data promises utilization of resources while ensuring accurate tumor predictions. In summary this study sheds light on how to enhance the methods of identifying tumors in MRI scans. It combines data processing, computer vision and deep learning techniques using the YOLOv8 model as a foundation. The discoveries made here have implications, for advancing medical image analysis. The goal is to achieve more tumor detection, which can greatly impact diagnosis and treatment processes.