BOTLCN: Improved Brain Tumor Detection and Classification by Transfer Learning Using Optimize Parameters with Butterfly Optimization

Abstract

The proposed research thus introduces a new approach utilizing Butterfly Optimization with Transfer Learning and Convolution Networks (BOTLCN) in the discovery and classification of cerebral tumors from medical imaging data. Rapidly increasing incidence and complexity of brain tumors call for sophisticated, optimal techniques for effectual diagnoses, which often fail to be furnished by conventional imaging technologies. The proposed BOTLCN model leverages the pattern recognition abilities of well-established Convolutional Neural Networks, like VGG-16, ResNet50, and DenseNet, to enhance analysis through sophisticated feature extraction.It therefore tries to optimize the model parameters with the butterfly optimization proposed as one of the nature-inspired algorithms that tune the parameters of the neural network to some function. To this extent, the algorithm fine-tunes the model over some predefined loss functions, like the Mean Squared Loss and Cross-Entropy, which are crucial in minimizing the diagnostic errors. Transfer learning is ensembled to mold the adopted approach around pre-existing neural architectures learned on diverse datasets to let convergence fast and feature robust extraction, which absolutely is paramount in the case of medical diagnostics.From the results of the work, the BOTLCN model showed better performance on the accuracy of traditional models at a percentage of 98.38%, a sensitivity of 97.33%, and a specificity of 99.10%. This attests to the model's ability to distinguish different types and grades of brain tumors with very high precision. This method improves the accuracy of tumor classification and gives detailed insights into the characteristics of the tumor for tailor-made treatment planning.In other words, the integration of state-of-the-art machine learning methodology into bio-inspired optimization, BOTLCN, is a significant milestone in the computational-based diagnosis of brain tumors which can serve as a tool to improve the outcome of clinical oncology.