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

Abstract

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.

Challenges of deep learning methods for COVID-19 detection using public datasets

“…Those features can be utilized to develop a sensitive Computer-aided Diagnosis (CAD) tool to detect COVID- Pneumonia and be considered as a screening tool [14] . Currently, deep Convolutional Neural Networks (CNNs) allow for building an end-to-end model, without the need for manual feature extraction [15] , [16] , which have demonstrated tremendous success in many domains of medical imaging, such as arrhythmia detection [17] , [18] , [19] , skin lesion segmentation and classification [20] , [21] , [22] , [23] , [24] , breast cancer detection [25] , [26] , [27] , brain disease classification [28] , pneumonia detection from CXR images [29] , fundus image segmentation [30] , [31] , minimally invasive surgery [32] and lung segmentation [33] . Several deep CNN-based methods have been published to detect COVID-19 from CXR and CT images.…”

Brain tumor classification by using a novel convolutional neural network structure

“…One of the strategies used to improve the performance in classification of medical signals or images is to use several models together 32–35 or the transfer learning 36,37 . It is also observed that this approach is used in the classification of brain tumors 26–28 .…”

A weakly supervised approach to skin lesion segmentation

“…CNNs are also becoming very popular in medical image analysis [7] and many decision support systems have been developed, for example, for automatic reporting of medical exams [8] and analysis of retinal fundus images [9]. Moreover, CNNs were also successfully employed in skin lesion analysis to classify and segment nevi and melanomas [10,11,12,13,14]. Melanoma is an aggressive form of cancer, triggered by an uncontrolled proliferation of melanocytes, pigment-producing cells of neuroectodermal origin.…”

“…Unfortunately, only a small fraction of the ISIC images was also labelled at the pixel-level, since the high cost of image tagging by medical experts made the collection of large sets of segmentation label maps very difficult. However, lesion segmentation also proved very useful in improving the classification accuracy [13,14], as the lesion boundaries are also used by dermatologists for their diagnosis. To this end, a weakly-supervised approach, inspired by [17], was employed to automatically extract segmentation supervisions for approximately 43,000 ISIC images.…”