Computing and Informatics

Computerized segmentation and classification of brain tumors are essential for diagnosis, management, and prediction of the prognosis in neuro-oncology. Standard methodologies for the diagnosis of brain tumors are subjective, based on human experience and qualitative analysis of images, which makes them unsuitable for handling all the variability in tumor types. To address these challenges, this research presents an innovative approach of combining DCNs and PECT images to accurately and quickly segment and classify multi-class brain tumor. The framework uses a DenseNet architecture tailored for high-resolution functional imaging in PECT to yield rich feature representations for tumor identification as well as grading and staging for several types of tumors. The computer algorithm was trained on PECT scans of 250 patients with gliomas, meningiomas and metastatic brain tumors as a training set and test set. Proper segmentation of tumor boundaries, type of tumor, and other important structures were carried out using the multi-class segmentation maps. By implication, the experimental results show that the proposed model has a classification accuracy of 94 %, which was higher than PECT-based approaches by 2 %. The generated segmentation maps also yield essential sparsely rendered images which really help the surgeon understand the tumour's size, position and relationship to adjacent structures and provide better planning and strategy in the treatment process. The framework was able to handle problems arising from the heterogeneity of tumor appearances and the spatial distribution in complicated neuro-oncology cases. This work underscores the role of the latest deep learning approaches in enhancing diagnostic accuracy and the ability to concurrently combine new imaging systems. In future studies, further work will be directed to the increased robustness achieved with the utilization of further imaging, enlarging the dataset for different demographics, and the testing of the proposed framework on more clinical scenarios.