Multiclass classification of brain tumors using a novel CNN architecture

Brain tumors are a deadly condition that radiologists have a tough time diagnosing. It is critical to make treatment-related decisions based on accurate and timely categorization of malignant cancers. Several approaches for detecting brain tumors have been presented in recent years. These strategies, however, necessitate handmade feature extraction and manual tumor segmentation prior to classification, which is error-prone and time-consuming. To properly extract features and identify brain cancers, an automated tumor diagnosis approach is necessary. Despite significant advancements in the development of such systems, the techniques face challenges due to low accuracy and large false-positive values. In this study, we propose a 13-layer CNN architecture for classifying brain tumors from MRI scans. We tested the suggested model's performance on a benchmark dataset of 3064 MRI images of three different types of brain cancer (glioma, pituitary, and meningioma) and achieved the highest accuracy of 97.2%, outperforming previous work on the same database. Furthermore, we validated our model on a cross-dataset scenario to demonstrate its efficacy in a real-world scenario. The main goal is to create a lightweight CNN architecture with fewer layers and learnable parameters that can reliably detect tumors in MRI images in the shortest amount of time. The findings show that the proposed technique is effective in classifying brain tumors using MRI images. Because of its adaptability, the proposed algorithm can be easily used in practice to assist doctors in diagnosing brain tumors at an early stage.