Breast Cancer Ultrasound Images Classification Using Hybrid Pre-Trained CNN and SVM

Breast cancer is simplest to treat while the lesion is tiny, but at that point there are no symptoms. As a result, screening is crucial for early detection. Many attempts have been made recently to reduce the operator dependency of Ultrasound (US) imaging and automate the diagnosing procedures to locate and categorize breast lesions, scientists used a range of algorithms and Computer-Aided Diagnosis (CAD) instrument. Classification is important to consider determining and detecting the type of cancer in images. The last CNN output layer is replaced with a preeminent classifier, such as support vector machines (SVM), can improve the pre-trained CNN architecture's classification accuracy. Here, seven leading deep learning pre-trained model for feature extraction such as VGG-19, VGG-16, ResNet-18, ResNet-50, EfficientNet-B0, MobileNet-V3, Dense Net and SVM are used for classification. 780 image samples from dataset BUSI with with 437 benign, 210 malignant, and 133 normal images. Each classifier's performance is evaluated and contrasted based on its accuracy. EfficientNetB0-SVM model had a best accuracy of 98.84 % in validation dataset and MobileNet-V3+SVM model had a best accuracy of $98,41 \%$ in testing dataset. The research results show that the hybrid model has the potential for greater accuracy to be used as a basis for medical image classification modeling in the future.