Deep convolutional neural network-based classification of cancer cells on cytological pleural effusion images

Abstract

Lung cancer is one of the leading causes of cancer-related death worldwide. Cytology plays an important role in the initial evaluation and diagnosis of patients with lung cancer. However, due to the subjectivity of cytopathologists and the region-dependent diagnostic levels, the low consistency of liquid-based cytological diagnosis results in certain proportions of misdiagnoses and missed diagnoses. In this study, we performed a weakly supervised deep learning method for the classification of benign and malignant cells in lung cytological images through a deep convolutional neural network (DCNN). A total of 404 cases of lung cancer cells in effusion cytology specimens from Shanghai Pulmonary Hospital were investigated, in which 266, 78, and 60 cases were used as the training, validation and test sets, respectively. The proposed method was evaluated on 60 whole-slide images (WSIs) of lung cancer pleural effusion specimens. This study showed that the method had an accuracy, sensitivity, and specificity respectively of 91.67%, 87.50% and 94.44% in classifying malignant and benign lesions (or normal). The area under the receiver operating characteristic (ROC) curve (AUC) was 0.9526 (95% confidence interval (CI): 0.9019-9.9909). In contrast, the average accuracies of senior and junior cytopathologists were 98.34% and 83.34%, respectively. The proposed deep learning method will be useful and may assist pathologists with different levels of experience in the diagnosis of cancer cells on cytological pleural effusion images in the future.

Fig. 1. Overview of the proposed deep learning framework presented in this study.

WSIs of lung cancer pleural effusion specimens were cropped into small patches and classified as benign or malignant lesions based on a Resnet18 deep convolutional neural network. WSI, whole-slide image.

Fig. 2. The consort diagram of the enrolled data from the Shanghai Pulmonary Hospital from March 2018 to January 2020.

Enrolled 404 WSIs were randomly allocated into training (266 WSIs), validation (78 WSIs) and testing (60 WSIs) datasets. WSI, whole slide image.

Fig. 3. Generation and augmentation of patch images.

Image patches were augmented by horizontal flip, vertical flip and color jitter.

Fig. 4. Receiver operator characteristics (ROC) curve of the AI model with AUC of 0.9526 (95% CI: 0.9019–0.9909).

AUC denotes the area under the receiver operator characteristics curve.

Fig. 5. Heatmap of Kendall correlation τ value for gold standard, two senior cytopathologists, two junior cytopathologists and DCNN model.

GS, Gold standard; P1 and P2, two senior pathologists; P3 and P4, two junior pathologists; AI, the Aitrox AI model based on a DCNN.

Fig. 6. Different situation of AI performance in pleura effusion cytology diagnosis.

a and b show AI made correct diagnosis of benign cells; c and d show AI made correct diagnosis of malignant cells; e and f show AI misdiagnosed benign mesothelial cells as malignant; g and h show AI misdiagnosed malignant cells as benign.