Fossa Optimization Algorithm with Convolution Neural Network Based Brain Tumor Segmentation and Classification in MRI Images

Magnetic Resonance Image (MRI) brain tumor segmentation is critical and significant in a clinical field that assist in diagnosis, complete growth prediction and tumor mass rates required for patients. The complexity in segmenting brain tumors are challenging due to the various structures, shapes, locations, visual characteristics, frequency, texture and contrast. To overcome this issue, this research proposes a Fossa Optimization Algorithm (FOA) with Convolution Neural Network (CNN) for brain tumor classification. The Geodesic Active Contour (GAC) is used for segmentation which identifies tumor boundaries precisely in noisy and intricate data. FOA utilizes dynamic search strategy by its hunting behavior, balancing exploration and exploitation effectively. Its adaptive step size enables better convergence when avoiding local optima. Furthermore, its ability to capture edges and patterns makes it effective for classification. CNN are better to diverse in scale, noise and rotation because of its shared weights and local connectivity reduce the complexity thereby leading to higher accuracy and quick processing in classification. The FOA-CNN obtains 99.92% accuracy, 99.20% f1-score, 99.07% sensitivity and 99.34% precision for BRATS2020 dataset which is superior than Spatial Transformer Network and Non-local Attention Mechanism (STN-NAM).