Transfer Learning Based Weighted Deep Learning Ensemble Model for Medical Image Classification

Abstract

Malignant melanoma is a well-known and deadly form of cancer that originates from epidermal melanocytes in humans. Early detection of such diseases, including various forms of cancer, is necessary for speeding up diagnosis and enhancing patient outcomes. A novel transfer learning-based ensemble-deep learning model was presented for diagnosing diseases at a preliminary stage. Data augmentation was used to increase the dataset, and integration of Inception-v3, DenseNet-121, and ResNet-50 techniques, along with an ensemble method, was employed to overcome the scarcity of labeled datasets and increase the accuracy as well as make the model more robust. The proposed system was trained and tested employing the International Skin Imaging Collaboration (ISIC) dataset. The suggested ensemble model gained the best performance, producing 98% accuracy, 98% area under the curve, 98% precision, and 98% F1 score. The proposed model outperformed the existing state-of-the-art models in disease classification. Furthermore, the proposed model will be beneficial for medical diagnosis and reduce the incidence of various diseases.