Although automated Skin Lesion Classification (SLC) is a crucial integral step in computer-aided diagnosis, it remains a challenging task due to inconsistency in texture, color, indistinguishable boundaries, and shapes. In this article, we propose an automatic and robust framework for the dermoscopic SLC named Dermoscopic Expert (DermoExpert). The DermoExpert consists of a preprocessing, a hybrid Convolutional Neural Network (hybrid-CNN), and transfer learning. The proposed hybrid-CNN classifier consists of three distinct feature extractors, with the same input images, which are fused to achieve better-depth feature maps of the corresponding lesion. Those distinct and fused feature maps are classified using the different fully connected layers, which are then ensembled to get a final prediction probability. In the preprocessing, we use lesion segmentation, augmentation, and class rebalancing. For boosting the lesion recognition, we have also employed geometric and intensity-based augmentation as well as the class rebalancing by penalizing the loss of the majority class and adding extra images to the minority classes. Additionally, we leverage the knowledge from a pre-trained model, also known as transfer learning, 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. We evaluate our DermoExpert on the ISIC-2016, ISIC-2017, and ISIC-2018 datasets, where our DermoExpert has achieved the area under the receiver operating characteristic curve (AUC) of 0.96, 0.95, and 0.97, respectively. The experimental results outperform the recent state-of-the-art by a margin of 10.0 % and 2.0 % respectively for ISIC-2016 and ISIC-2017 datasets in terms of AUC. The DermoExpert also outperforms, in concerning a balanced accuracy, by a margin of 3.0 % for ISIC-2018 dataset. Since our framework can provide better-classification on three different test datasets, even with limited training data, it can lead to better-recognition of melanoma to aid dermatologists. Our source code, and segmented masks, for ISIC-2018 dataset, will be made publicly available for the research community for further improvements.

A large number of studies in the past months have proposed deep learning-based Artificial Intelligence (AI) tools for automated detection of COVID-19 using publicly available datasets of Chest X-rays (CXRs) or CT scans for training and evaluation. Most of these studies report high accuracy when classifying COVID-19 patients from normal or other commonly occurring pneumonia cases. However, these results are often obtained on cross-validation studies without an independent test set coming from a separate dataset and have biases such as the two classes to be predicted come from two completely different datasets. In this work, we investigate potential overfitting and biases in such studies by designing different experimental setups within the available public data constraints and highlight the challenges and limitations of developing deep learning models with such datasets. We propose a deep learning architecture for COVID-19 classification that combines two very popular classification networks, ResNet and Xception, and use it to carry out the experiments to investigate challenges and limitations. The results show that the deep learning models can overestimate their performance due to biases in the experimental design and overfitting to the training dataset. We compare the proposed architecture to state-of-the-art methods utilizing an independent test set for evaluation, where some of the identified bias and overfitting issues are reduced. Although our proposed deep learning architecture gives the best performance with our best possible setup, we highlight the challenges in comparing and interpreting various deep learning algorithms’ results. While the deep learning-based methods using chest imaging data show promise in being helpful for clinical management and triage of COVID-19 patients, our experiments suggest that a larger, more comprehensive database with less bias is necessary for developing tools applicable in real clinical settings.