Deep Learning-Based Hepatocellular Carcinoma Histopathology Image Classification: Accuracy Versus Training Dataset Size

Globally, liver cancer causes more than 700,000 deaths each year and is the second-leading cause of death from cancer. Hepatocellular carcinoma (HCC) is the most common type of liver cancer in adults and accounts for most deaths in cirrhosis patients. Patients with early-stage liver cancer can be treated by surgical intervention with a good prognosis; thus, early diagnosis, as confirmed by liver pathology examination, is necessary to combat HCC. Conventional manual pathology examination requires considerable time and labor, even with established expertise. It is widely accepted that intelligent classifiers may prove effective in the diagnosis process. In this study, we used a GoogLeNet (Inception-V1)-based binary classifier to classify HCC histopathology images. The classifier achieved 91.37% (+/- 2.49) accuracy, 92.16% (+/- 4.93) sensitivity, and 90.57% (+/- 2.54) specificity in HCC classification. Although the classification accuracy of deep learning is reported to be positively correlated with the amount of training data, it is often uncertain how much training data are required for deep learning to achieve satisfactory performance in clinical diagnosis. Moreover, deep learning methods require annotated data to generate efficient models. However, annotated data are a relatively scarce resource and can be expensive to obtain. Hence, the relationship between classification accuracy and the number of liver histopathology images for training was investigated. An inverse power law function-based estimation model is proposed to evaluate the minimum number of annotated training images required for a desired diagnostic accuracy.