DermoExpert: Skin lesion classification using a hybrid convolutional neural network through segmentation, transfer learning, and augmentation

Background and Objective Although automated Skin Lesion Classification (SLC) is a crucial integral step in computeraided diagnosis, it remains challenging due to inconsistency in textures, colors, indistinguishable boundaries, and shapes.

Methods This article proposes an automated dermoscopic SLC framework named Dermoscopic Expert (DermoExpert). The DermoExpert consists of preprocessing and hybrid Convolutional Neural Network (hybrid-CNN), leveraging a transfer learning strategy. The proposed hybrid-CNN classifier has three different feature extractor modules taking the same input images, which are fused to achieve better-depth feature maps of the corresponding lesion. Those unique and fused feature maps are classified using different fully connected layers, which are then ensembled to predict the lesion class. We apply lesion segmentation, augmentation, and class rebalancing in the proposed preprocessing. We have also employed geometry- and intensity-based augmentations and class rebalancing by penalizing the majority class’s loss and combining additional images to the minority classes to enhance lesion recognition outcomes. Moreover, we leverage the knowledge from a pre-trained model to build a generic classifier, although small datasets are being used. In the end, we design and implement a web application by deploying the weights of our DermoExpert for automatic lesion recognition.

Results We evaluate our DermoExpert on the ISIC-2016, ISIC-2017, and ISIC-2018 datasets, where the DermoExpert has achieved the area under the receiver operating characteristic curve (AUC) of 0.96, 0.95, and 0.97, respectively. The experimental results defeat the recent state-of-the-art by the margins of 10.0 % and 2.0 % respectively for the ISIC-2016 and ISIC-2017 datasets in terms of AUC. The DermoExpert also outperforms by a border of 3.0 % for the ISIC-2018 dataset concerning a balanced accuracy.

Conclusion Since our framework can provide better-classification outcomes on three different test datasets, it can lead to better-recognition of melanoma to assist dermatologists. Our source code and segmented masks for the ISIC-2018 dataset will be publicly available for further improvements.