Two-Step Approach for Brain Tumor Classification using Deep Learning Technique

The paper presents a two-stage deep learning method for detecting the brain tumor and classifying the type of brain tumors using MRI scan. The Capsule Neural Network model checks, in the first stage, whether a tumor is present or not by making use of an MRI dataset labeled with regards to tumor or no-tumor. Now, if it is a tumor, then the second stage classifies the tumors according to 16 categories based on a CNN model with a Swin Transformer. The same CNN, now enhanced with transfer learning using a pre-trained Swin Transformer model, will compute confidence scores for all categories, being more certain about the model's predictions. These confidence scores are of importance because they provide more than the most likely type of tumor: how confident the model is with the decision. The testing accuracy achieved by the CapsNet model was 97.83%, while the CNN based on Swin Transformer reached an overall accuracy of 92.86 %. Thorough testing has shown both models to perform very well in their tasks. In that respect, such a two-step process is greatly improved compared to the traditional diagnosis ways and offers adequate potential for accurately diagnosing the brain tumors. It is combined with CapsNet and CNN enhanced Swin-Transformer by transfer learning. This makes the method more efficient in terms of diagnostic accuracy.