An efficient multi-functional deep learning model for effective medical image classification using skin lesion database

Kishore Babu Nampalle, Balasubramanian Raman

Abstract: The automatic process of classifying a medical image plays a vital role in Computer-Aided Diagnosis (CADx). Due to the advent of Convolutional Neural Networks (CNNs) and wide usage, there has been a substantial improvement in the performance of the classification process combined with the process of implicit feature extraction. CNN requires a large amount of data, but building an extensive data set is challenging. Hence, Transfer learning appeared to resolve the same issue. Predefined models like MobileNet, VGG19, Inception-V3, and ResNet50, based on datasets with more sizes such as ImageNet, playa vital role in training and improving the performance. Extracting such unique features from medical images is a challenging task due to the different properties of images. Training a Deep Neural Network is an intensive task because it requires high configured computing machines and may require more time. Hence, this paper proposed a multi-functional deep learning architecture, including an ensemble of Logistic Regression classifiers and a MobileNet pre-trained model. Here, the input data of skin lesion images from the ISI C challenge dataset for binary and multi-class classification. Obtained results are compared with other models with the help of performance metrics.