Hybrid deep-cuckoo framework for robust brain tumor detection and segmentation in MRI scans

Doctors detect brain tumor manually through MRI images, which involves subjective judgements and is time consuming. For better brain tumor detection, this research proposes a hybrid model that combines the ability to learn features of CNNs with the Cuckoo Search Algorithm which improves the CNN performance and makes the training efficient. Many experiments are done using publicly accessible datasets such as Figshare and BraTS brain MRI datasets, with tasks focused on classifying three types of tumors (glioma, meningioma and pituitary tumor) and segmentation. Classifying images and segmenting pixels precisely are two tasks the proposed model undertakes successfully and more efficiently than CNN in terms of accuracy of 96.7%, precision of 94.0%, recall of 93.1%, and F1-score of 96.5%. The study also uses Grad-CAM to show how the model decides and makes its predictions, helping doctors and nurses feel more comfortable and understanding.