The Role of Artificial Intelligence With Deep Convolutional Neural Network in Screening Melanoma: A Systematic Review and Meta-Analyses of Quasi-Experimental Diagnostic Studies

Abstract

Introduction: Detecting melanoma as one of the most common skin cancer with using artificial intelligence (AI), such as deep convolutional neural network (DCNN) have the potency to increase the accuracy of the diagnosis. The aim of this study is to analyze the sensitivity, specificity, precision, and F1-score of DCNN in screening melanoma. Methodology: The authors followed the PRISMA 2020 guidelines to retrieve literature in the following databases: PubMed, EBSCOhost, Emerald, Wiley, and ScienceDirect. The study’s inclusion criteria were human quasi-experimental investigated DCNN in screening melanoma. The analysis was conducted using RevMan 5.4 and Quality Assessment of Diagnostic Accuracy Studies 2 (QUADAS-2) to ensure the quality of the studies. Results: Fifty-six of 2386 articles published in 2003 to 2023 were included and 24 studies were statistically analyzed. Various type of DCNN was used [artificial neural network (n = 4); pigment network (n = 4); atypical pigment network (n = 1); ResNet (= 8); AlexNet (n = 3); visual geometry group (n = 7); inception (n = 4); custom DCNN (n = 4)]. The mean and median of total sample size in meta-analysis with melanoma subjects were (18,791; 2,157) with (573; 261), respectively. Overall, QUADAS-2 showed low risk of bias. Diagnostic performance was observed with pooled sensitivity (0.881), pooled specificity (0.897), and pooled AUC (0.894). The precision and F1-score were ranging from 58% to 98.83% and 0.45 to 0.98. The forest plot and summary receiver operating characteristics curve (SROC) of each multiple in multiple analysis showed satisfactory results. Conclusions: DCNN showed significant result to screen melanoma in patients. It has the potential to help clinician in giving early screening.