Automated Diagnosis and Localization of Melanoma from Skin Histopathology Slides Using Deep Learning: A Multicenter Study

Abstract

In traditional hospital systems, diagnosis and localization of melanoma are the critical challenges for pathological analysis, treatment instructions, and prognosis evaluation particularly in skin diseases. In literature, various studies have been reported to address these issues; however, a prominent smart diagnosis system is needed to be developed for the smart healthcare system. In this study, a deep learning-enabled diagnostic system is proposed and implemented that it has the capacity to automatically detect malignant melanoma in whole slide images (WSIs). In this system, the convolutional neural network (CNN), sophisticated statistical method, and image processing algorithms were integrated and implemented to locate benign and malignant lesions which are extremely useful in the diagnoses process of melanoma disease. To verify the exceptional performance of the proposed scheme, it is implemented in a multicenter database, which has 701 WSIs (641 WSIs from Central South University Xiangya Hospital (CSUXH) and 60 WSIs from the Cancer Genome Atlas (TCGA)). Experimental results have verified that the proposed system has achieved an area under the receiver operating characteristic curve (AUROC) of 0.971. Furthermore, the lesion area on the WSIs is represented by its degree of malignancy. These results show that the proposed system has the capacity to fully automate the diagnosis and localization problem of the melanoma in the smart healthcare systems.

Deep learning for the detection of anatomical tissue structures and neoplasms of the skin on scanned histopathological tissue sections

“…In this regard, a higher number of patient samples is commonly preferred over a large number of images. The number of patients included for training, validation and testing in the current investigation is within the reported range of previous studies ( 9 , 11 , 26 ). Currently, there is no consensus on the minimum number of cases that should be included in a deep learning study.…”

“…In skin diseases, the technique has been mainly used in neoplastic diseases so far. While most reports focus on either melanocytic ( 9 – 11 , 30 ) or non-melanocytic ( 26 ) lesions, data on both melanocytic and non-melanocytic lesions, non-tumor skin lesions or anatomical tissue structures are scarce ( 12 , 26 ). Although there are honorable exceptions ( 31 ), most of the published studies on deep learning on histopathological slides do not make their annotated data, the full dataset of image patches and/or their code available.…”

“…The use of deep learning methods applied to clinical images, dermatoscopy images or scanned histopathological slides holds great promise to support cancer diagnostics in general ( 7 , 8 ), and skin cancer diagnostics in particular ( 5 , 9 ). In the past, the feasibility and potential to classify different diseases on scanned histological slides has been demonstrated for automated localization and diagnosis of melanoma ( 9 , 10 ), the differentiation between naevi and melanoma ( 11 ), and the differentiation between basaloid, squamous, melanocytic and other skin tumors ( 12 ).…”

Development and Validation of an Artificial Intelligence-Based Image Classification Method for Pathological Diagnosis in Patients With Extramammary Paget’s Disease

“…Li et al. proposed a deep learning-based pathology diagnosis system for melanoma whole slide imaging (WSI) classification and generated a multicenter WSI database for model training, which could assist the pathological diagnosis of melanoma diseases ( 11 ). Xie et al.…”

“…Thomas et al found that some of the most common skin cancer diagnoses, such as basal cell carcinoma (BCC), squamous cell carcinoma (SCC), and intraepidermal carcinoma (IEC) are amenable to deep learning methods (10). Li et al proposed a deep learning-based pathology diagnosis system for melanoma whole slide imaging (WSI) classification and generated a multicenter WSI database for model training, which could assist the pathological diagnosis of melanoma diseases (11). Xie et al collected 2,241 digital whole-slide images from 1,321 patients and constructed a multicenter dataset for training both ResNet50 and Vgg19 to test performance with the classification of melanoma and nevi.…”