Application of a hybrid EfficientNet-SVM model to medical image classification

Within the realm of medical science, imaging stands out as the most potent diagnostic and therapeutic tool. Convolutional Neural Networks (CNN) have revolutionized this field due to their ability to extract complex features and learn intricate patterns from raw image data. However, due to the complexity of medical image data and the growing demand for greater precision, CNNs struggle to operate at their best within realistic computing timeframes. In response to these challenges, the hybridization of CNN with classification algorithms such as SVM has ameliorated its performance. However, additional improvements are required to bring it into compliance with the requirements of medical image analysis. This study introduces an innovative methodology that leverages a contemporary metaheuristic algorithm to augment the efficiency and efficacy of CNN-SVM within medical image analysis tasks, employing an EfficientNet architecture. Brain Tumor Database (BTD); an MRI-associated dataset was used in this research. In the suggested approach, automatic feature extraction from images is done using CNN, and the final classification is done with Support Vector Machines (SVM). The evaluation of the suggested model entails two categories of statistical assessment metrics: Accuracy and F-measure. The overall accuracy achieved by the proposed model stands at 95.45% for the Brain Tumor Database (BTD).