Brain Tumor Detection Using Optimization with Genetic and Heuristic Algorithms

Brain tumor MRI imaging is an important procedure in clinical judgment and primary diagnosis. Analyzing MRI scans manually requires significant time and may lead to diagnostic inaccuracies; hence, automated and intelligent systems should be applied in the classification of the scans in order to get reliable results. The proposed methodology is a refinement of a useful deep learning network to identify brain MRI tumors using the EfficientNetB3 architecture, pertained in two phases of training. There are 4 types of data: No Tumor, Glioma, pituitary and Meningioma are the preprocessing technique used is strong i.e. auto- cropping, histogram equalization and normalization are applied that enhance image quality and make them more consistent. The EfficientNetB3 base layers had initially been frozen to extract features and subsequently unfrozen in selective mode to optimize the results of the fine-tuning process in such a way that the model can form domain-specific features with this sort of efficiency. There is a general accuracy of approximately 98 percent with high sensitivity and specificity of all tumor classes in factors of experimental results, which are better obtained. The solid nature of the model and its generalization power is demonstrated through the measurement variables including the confusion matrix, the precision, the recall and F1-score. This was also accompanied by the design of a lightweight Colab runtime interface which was able to predict and visualize images in real- time and was able to be easily deployed both in clinical and research settings. This proposed System is based on a CNN architecture integrated with transfer learning.