Automatic volumetric diagnosis of hepatocellular carcinoma based on four-phase CT scans with minimum extra information - PubMed (original) (raw)
Automatic volumetric diagnosis of hepatocellular carcinoma based on four-phase CT scans with minimum extra information
Yating Ling et al. Front Oncol. 2022.
Abstract
Summary: We built a deep-learning based model for diagnosis of HCC with typical images from four-phase CT and MEI, demonstrating high performance and excellent efficiency.
Objectives: The aim of this study was to develop a deep-learning-based model for the diagnosis of hepatocellular carcinoma.
Materials and methods: This clinical retrospective study uses CT scans of liver tumors over four phases (non-enhanced phase, arterial phase, portal venous phase, and delayed phase). Tumors were diagnosed as hepatocellular carcinoma (HCC) and non-hepatocellular carcinoma (non-HCC) including cyst, hemangioma (HA), and intrahepatic cholangiocarcinoma (ICC). A total of 601 liver lesions from 479 patients (56 years ± 11 [standard deviation]; 350 men) are evaluated between 2014 and 2017 for a total of 315 HCCs and 286 non-HCCs including 64 cysts, 178 HAs, and 44 ICCs. A total of 481 liver lesions were randomly assigned to the training set, and the remaining 120 liver lesions constituted the validation set. A deep learning model using 3D convolutional neural network (CNN) and multilayer perceptron is trained based on CT scans and minimum extra information (MEI) including text input of patient age and gender as well as automatically extracted lesion location and size from image data. Fivefold cross-validations were performed using randomly split datasets. Diagnosis accuracy and efficiency of the trained model were compared with that of the radiologists using a validation set on which the model showed matched performance to the fivefold average. Student's _t_-test (T-test) of accuracy between the model and the two radiologists was performed.
Results: The accuracy for diagnosing HCCs of the proposed model was 94.17% (113 of 120), significantly higher than those of the radiologists, being 90.83% (109 of 120, _p_-value = 0.018) and 83.33% (100 of 120, _p_-value = 0.002). The average time analyzing each lesion by our proposed model on one Graphics Processing Unit was 0.13 s, which was about 250 times faster than that of the two radiologists who needed, on average, 30 s and 37.5 s instead.
Conclusion: The proposed model trained on a few hundred samples with MEI demonstrates a diagnostic accuracy significantly higher than the two radiologists with a classification runtime about 250 times faster than that of the two radiologists and therefore could be easily incorporated into the clinical workflow to dramatically reduce the workload of radiologists.
Keywords: arificial intelligence; computed tomography; deep learning; diagnosis; hepatocellular carcinoma.
Copyright © 2022 Ling, Ying, Xu, Peng, Mao, Chen, Ni, Liu, Gong and Kong.
Conflict of interest statement
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Figures
Figure 1
Flowchart of our study. Participant selection, model training, model testing and reader study are included in our study.
Figure 2
Overview of the proposed method. The upper part is MEI pathway and the lower part is the CT pathway. The 3D ResNet in CT pathway contains 14 layers (13 convolution layers, and 1 global average pooling layer). Filter size of the first convolution layer is 5×5×5, and the following filter sizes are 3×3×3. Filter size of the global average pooling layer is 2×2×2. The basic model and enhanced model only have the CT pathway. The size of image input in basic model is 64×64×64×1 while the others are 64×64×64×4. MEI, Minimum Extra Information.
Figure 3
ROC curves of basic model and enhanced model. The lines reflect the average performances of the models, and the light-colored area reflects the fluctuation of the models represented by the corresponding standard deviations.
Figure 4
The liver masses misdiagnosed by models. (A) shows four phase images of a 62-year-old man with a hemangioma (arrow) that was diagnosed through one-year follow-up in 2018. The mass was correctly diagnosed as non-HCC by using enhanced model and our MExPale model. It was misdiagnosed as HCC by using basic model. (B) shows four phase images of 54-year-old man with a HCC (arrow) that was diagnosed after surgery. The mass was correctly diagnosed as HCC by using our MExPale model. It was misdiagnosed as HCC by using basic model and enhanced model.
Figure 5
The average accuracy and standard deviations of different models. Model 1, Enhanced model; Model 2, Enhanced model with spatial morphological information; Model 3, Enhanced model with spatial morphological information and age; Model 4, Enhanced model with morphological information and gender; Model 5, MExPale model.
Figure 6
Performance of models. (A) ROC curves of models, (B) The confusion matrix of our MExPale model. HCC, hepatocellular carcinoma.
Figure 7
ROC curves of our MExPale model and two radiologists. Radiologist 1 comes from the First Affiliated Hospital of Zhejiang University, and Radiologist 2 comes from a community primary hospital.
Figure 8
The liver masses misdiagnosed by model and two radiologists. (A) shows four phase images of a 59-year-old man with a HCC (arrow) that was diagnosed after surgery. The mass was misdiagnosed diagnosed as non-HCC by and our MExPale model and both two radiologists. (B) shows four images of a 64-year-old man with a ICC (arrow) that was diagnosed after surgery. The mass was misdiagnosed diagnosed as HCC by our MExPale model and both two radiologists.
References
LinkOut - more resources
Full Text Sources
Other Literature Sources