Hybrid fractal ConvXGBNet for brain tumor classification with log-cosh dice loss - cascaded V-Net based segmentation based on MRI images.

As one of the body's most vital organs, the brain oversees a wide array of functions, including cognition, memory, motor skills, and sensory processing. Brain tumors occur when cells in the brain grow uncontrollably, forming a mass or lump that may be either cancerous or non-cancerous. Timely detection is key to effective treatment, as tumors can interfere with how the brain works and lead to major health issues. Traditional methods faced challenges like human error, long processing times, and limited accuracy. An automatic brain tumor classification model has been created to tackle these issues, offering efficient and trustworthy diagnostic results in medical settings where time is critical. Here, a Fractal Convolutional eXtreme Gradient Boosting (Fractal-ConvXGBNet) framework is projected for the classification of brain tumor using Magnetic Resonance Imaging (MRI). The process begins with obtaining MRI brain images from databases, followed by applying high-boost filtering to refine and enhance image clarity. Subsequently, brain tumor segmentation is carried out via the Cascaded V-Net model, which is refined using a Log-Cosh Dice Loss-based loss function to achieve higher segmentation accuracy. After segmentation, feature extraction is performed using the Spatial Grey-Level Dependence Matrix (SGLDM) method. The final step involves brain tumor classification through the Fractal-ConvXGBNet model, a novel hybrid architecture that combines FractalNet with ConvXGB to improve classification accuracy. The developed model classifies brain MRI images into four distinct categories: Glioma, Meningioma, Pituitary tumor, and Normal. Subsequently, the Fractal-ConvXGBNet achieved an accuracy of about 90.75%, True Negative Rate (TNR) of 91.76% and True Positive Rate (TPR) of 89.89%.