Artificial intelligence in histologic diagnosis of ductal carcinoma in situ

Abstract

<p>Image analysis is key to diagnostic accuracy, and in the era of big data, artificial intelligence (AI) can be applied to improve both the accuracy and efficiency of diagnoses. Traditional histopathological diagnosis of breast ductal carcinoma in situ (DCIS) remains difficult due to the strong similarities between its histopathological features and those of atypical ductal hyperplasia (ADH). Recent advancements in AI have enabled its use in histopathological image analysis, largely in the form of convolutional neural networks (CNNs) and support vector machines (SVMs). This is a systematic review conducted in line with the PRISMA protocol. Four databases were searched on 7 September 2021: Embase, PubMed, Scopus, and Web of Science. 23 studies were included. The following parameters were extracted: the accuracy, sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) of each study. Analysis of the included studies revealed that AI-assisted histopathological analysis is of high accuracy (83.78%), sensitivity (83.88%), and specificity (85.49%), and has a high PPV (89.43%). Our results also showed that CNN is the most commonly used mode of machine learning – 21 models used only CNN, while 2 models used only SVM. On average, CNN yielded slightly higher accuracy and sensitivity (86.71% and 85.22% respectively) than SVM (accuracy: 85.00%; sensitivity: 70.00%). When the two methods were combined, a comparably high mean accuracy of 82.52% and a mean sensitivity of 83.00% were achieved. The use of AI as a diagnostic adjunct can significantly improve the accuracy and efficiency of DCIS diagnosis and can therefore reduce pathologists’ workload.</p>