Dataset Information

Development of an AI system for accurately diagnose hepatocellular carcinoma from computed tomography imaging data.

ABSTRACT:

Background and aims

Computed tomography (CT) scan is frequently used to detect hepatocellular carcinoma (HCC) in routine clinical practice. The aim of this study is to develop a deep-learning AI system to improve the diagnostic accuracy of HCC by analysing liver CT imaging data.

Methods

We developed a deep-learning AI system by training on CT images from 7512 patients at Henan Provincial Peoples' Hospital. Its performance was validated on one internal test set (Henan Provincial Peoples' Hospital, n = 385) and one external test set (Henan Provincial Cancer Hospital, n = 556). The area under the receiver-operating characteristic curve (AUROC) was used as the primary classification metric. Accuracy, sensitivity, specificity, precision, negative predictive value and F1 metric were used to measure the performance of AI systems and radiologists.

Results

AI system achieved high performance in identifying HCC patients, with AUROC of 0.887 (95% CI 0.855-0.919) on the internal test set and 0.883 (95% CI 0.855-0.911) on the external test set. For internal test set, accuracy was 81.0% (76.8-84.8%), sensitivity was 78.4% (72.4-83.7%), specificity was 84.4% (78.0-89.6%) and F1 (harmonic average of precision and recall rate) was 0.824. For external test set, accuracy was 81.3% (77.8-84.5%), sensitivity was 89.4% (85.0-92.8%), specificity was 74.0% (68.5-78.9%) and F1 was 0.819. Compared with radiologists, AI system achieved comparable accuracy and F1 metric on internal test set (0.853 versus 0.818, P = 0.107; 0.863 vs. 0.824, P = 0.082) and external test set (0.805 vs. 0.793, P = 0.663; 0.810 vs. 0.814, P = 0.866). The predicted HCC risk scores by AI system in HCC patients with multiple tumours and high fibrosis stage were higher than those with solitary tumour and low fibrosis stage (tumour number: 0.197 vs. 0.138, P = 0.006; fibrosis stage: 0.183 vs. 0.127, P < 0.001). Radiologists' review showed that the accuracy of saliency heatmaps predicted by algorithms was 92.1% (95% CI: 89.2-95.0%).

Conclusions

AI system achieved high performance in the detection of HCC compared with a group of specialised radiologists. Further investigation by prospective clinical trials was necessitated to verify this model.

SUBMITTER: Wang M

PROVIDER: S-EPMC8505513 | biostudies-literature | 2021 Oct

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

Development of an AI system for accurately diagnose hepatocellular carcinoma from computed tomography imaging data.

Wang Meiyun M Fu Fangfang F Zheng Bingjie B Bai Yan Y Wu Qingxia Q Wu Jianqiang J Sun Lin L Liu Qiuyu Q Liu Mingge M Yang Yichen Y Shen Hongru H Kong Dalu D Ma Xiaoyue X You Peiting P Li Xiangchun X Tian Fei F

British journal of cancer 20210807 8

<h4>Background and aims</h4>Computed tomography (CT) scan is frequently used to detect hepatocellular carcinoma (HCC) in routine clinical practice. The aim of this study is to develop a deep-learning AI system to improve the diagnostic accuracy of HCC by analysing liver CT imaging data.<h4>Methods</h4>We developed a deep-learning AI system by training on CT images from 7512 patients at Henan Provincial Peoples' Hospital. Its performance was validated on one internal test set (Henan Provincial Pe ...[more]

PMID: 34365472

Similar Datasets

Project description:De novo lipogenesis is upregulated in many cancers, and targeting it represents a metabolic approach to cancer treatment. However, the treatment response is unpredictable because lipogenic activity varies greatly among individual tumors, thereby necessitating the assessment of lipogenic activity before treatment. Here, we proposed an imaging probe, positron emission tomography/computed tomography (PET/CT) with dual tracers combining 11C-acetate and 18F-fluorodeoxyglucose (18F-FDG), to assess the lipogenic activity of hepatocellular carcinoma (HCC) and predict the response to lipogenesis-targeted therapy. We investigated the association between 11C-acetate/18F-FDG uptake and de novo lipogenesis in three HCC cell lines (from well-differentiated to poorly differentiated: HepG2, Hep3B, and SkHep1) by examining the expression of lipogenic enzymes: acetyl-CoA synthetase 2 (ACSS2), fatty acid synthase (FASN), and ATP citrate lyase (ACLY). The glycolysis level was determined through glycolytic enzymes: pyruvate dehydrogenase expression (PDH). On the basis of the findings of dual-tracer PET/CT, we evaluated the treatment response to a lipase inhibitor (orlistat) in cell culture experiments and xenograft mice. Dual-tracer PET/CT revealed the lipogenic activity of various HCC cells, which was positively associated with 11C-acetate uptake and negatively associated with 18F-FDG uptake. This finding represents the negative association between 11C-acetate and 18F-FDG uptake. Because these two tracers revealed the lipogenic and glycolytic activity, respectively, which implies an antagonism between lipogenic metabolism and glucose metabolism in HCC. In addition, dual-tracer PET/CT not only revealed the lipogenic activity but also predicted the treatment response to lipogenesis-targeted therapy. For example, HepG2 xenografts with high 11C-acetate but low 18F-FDG uptake exhibited high lipogenic activity and responded well to orlistat treatment, whereas SkHep1 xenografts with low 11C-acetate but high 18F-FDG uptake exhibited lower lipogenic activity and poor response to orlistat. The proposed non-invasive dual-tracer PET/CT imaging can reveal the lipogenesis and glycolysis status of HCC, thus providing an ideal imaging probe for predicting the therapeutic response of HCC to lipogenesis-targeted therapy.

Project description:Background & aimsClinically significant portal hypertension (CSPH) is a landmark in the natural history of cirrhosis, influencing clinical decisions in patients with hepatocellular carcinoma (HCC). Previous small series suggested that splanchnic volume measurements may predict portal hypertension. We aimed to evaluate whether volumetry obtained by standard multidetector computerised tomography (MDCT) can predict CSPH in patients with HCC.MethodsWe included 175 patients with HCC, referred for hepatic venous pressure gradient (HVPG) evaluation, in whom contemporary MDCT was available. Liver volume, spleen volume (SV) and liver segmental volume ratio (LSVR: volume of the segments I-III/volume of the segments IV-VIII) were calculated semi-automatically from MDCT. Other non-invasive tests (NITs) were also employed.ResultsVolume parameters could be measured in almost 100% of cases with an excellent inter-observer agreement (intraclass correlation coefficient >0.950). SV and LSVR were independently associated with CSPH (HVPG ≥10 mmHg) and did not interact with aetiology. The volume Index (VI), calculated as the product of SV and LSVR, predicted CSPH (AUC 0.83; 95% CI 0.77-0.89). Similar results were observed in an external cohort (n = 23) (AUC 0.87; 95% CI 0.69-1.00). Setting a sensitivity and specificity of 98%, VI could have avoided 35.9% of HVPG measurements. The accuracy of VI was similar to that of other NITs. VI also accurately predicted HVPG greater than 12, 14, 16 and 18 mmHg (AUC 0.81 [95% CI 0.74-0.88], 0.84 [95% CI 0.77-0.91], 0.85 [95% CI 0.77-0.92] and 0.87 [95% CI 0.79-0.94], respectively).ConclusionsQuantification of liver and spleen volumes by MDCT is a simple, accurate and reliable method of CSPH estimation in patients with compensated cirrhosis and HCC.Impact and implicationsAn increase in portal pressure strongly impacts outcomes after surgery in patients with early hepatocellular carcinoma (HCC). Direct measurement through hepatic vein catheterization remains the reference standard for portal pressure assessment, but its invasiveness limits its application. Therefore, we evaluated the ability of CT scan-based liver and spleen volume measurements to predict portal hypertension in patients with HCC. Our results indicate that the newly described index, based on quantification of liver and spleen volume, accurately predicts portal hypertension. These results suggest that a single imaging test may be used to diagnose and stage HCC, while providing an accurate estimation of portal hypertension, thus helping to stratify surgical risks.

Project description:BackgroundDue to the high recurrence rate in hepatocellular carcinoma (HCC) after resection, preoperative prognostic prediction of HCC is important for appropriate patient management. Exploring and developing preoperative diagnostic methods has great clinical value in treating patients with HCC. This study sought to develop and evaluate a novel combined clinical predictive model based on standard triphasic computed tomography (CT) to discriminate microvascular invasion (MVI) in hepatocellular carcinoma (HCC).MethodsThe preoperative findings of 82 patients with HCC, including conventional clinical factors, CT imaging findings, and CT texture analysis (TA), were analyzed retrospectively. All included cases were divided into MVI-negative (n = 33; no MVI) and MVI-positive (n = 49; low or high risk of MVI) groups. TA parameters were extracted from non-enhanced, arterial, portal venous, and equilibrium phase images and subsequently calculated using the Artificial Intelligence Kit. After statistical analyses, a clinical model comprising conventional clinical and CT image risk factors, radiomics signature models, and a novel combined model (fused radiomic signature) was constructed. The area under the curve (AUC) of the receiver operating characteristics (ROC) curve was used to assess the performance of the various models in discriminating MVI.ResultsWe found that tumor diameter and pathological grade were effective clinical predictors in clinical model and 12 radiomics features were effective for MVI prediction of each CT phase. The AUCs of the clinical, plain, artery, venous, and delay models were 0.77 (95% CI: 0.67-0.88), 0.75 (95% CI: 0.64-0.87), 0.79 (95% CI: 0.69-0.89), 0.73 (95% CI: 0.61-0.85), and 0.80 (95% CI: 0.70-0.91), respectively. The novel combined model exhibited the best performance, with an AUC of 0.83 (95% CI: 0.74-0.93).ConclusionsModels derived from triphasic CT can preoperatively predict MVI in patients with HCC. Of the models tested here, the novel combined model was most predictive and could become a useful tool to guide subsequent personalized treatment of HCC.

Project description:BackgroundThe Ki-67 expression level, which represents the proliferation status of cells, serves as a prognostic marker for hepatocellular carcinoma (HCC); however, the immunochemistry method currently used to evaluate Ki-67 is invasive and is not suitable for patients who have lost the chance for surgery, such as those with advanced tumors or those with other serious organic diseases. This study aimed to investigate the value of quantitative dual-energy computed tomography (DECT) parameters in predicting the expression level of Ki-67 in HCC.MethodsThis study analyzed the data of 123 consecutive patients with HCC who underwent both Ki-67 immunohistochemistry analysis and two-phase contrast-enhanced DECT imaging. The patients were divided into the following two groups based on the positive rate of Ki-67 (Ki-67%): the high-expression group (Ki-67% >20%, n=74); and the low-expression group (Ki-67% ≤20%, n=49). The computed tomography (CT) values in the 130 and 140 keV monochromatic energy images (HU130-140keV), normalized effective atomic number (NeffZ), fat density (Dfat), and water density (Dwater) were measured and calculated at the arterial phase (AP) and portal venous phase (PVP). A Spearman correlation coefficient analysis, a comparison of parameters between groups, a receiver operating characteristic (ROC) curve analysis for evaluating predictive efficacy, and a multivariable logistic regression analysis were conducted.ResultsThe NeffZ-AP, HU130 keV-PVP, HU140 keV-PVP, Dfat-PVP, and Dwater-PVP were positively correlated with the Ki-67% (all P<0.05), and the DECT parameter values in the high-expression group were significantly higher than those in the low-expression group (all P<0.05). The Dwater-PVP [odds ratio (OR) =1.353, 95% confidence interval (CI): 1.204-1.521, P<0.001] and tumor diameter (OR =1.258, 95% CI: 1.08-1.465, P=0.003) were independent predictive factors for high Ki-67 expression. Dwater-PVP had the highest predictive efficacy with an area under the curve (AUC) of 0.810. The multivariable analysis combining the DECT parameters and morphological characteristics improved the predictive efficacy of the model in predicting high Ki-67 expression (AUC =0.857).ConclusionsDECT provides a non-invasive method to evaluate the proliferation status of HCC cells, and the predictive efficacy of high Ki-67 expression could be improved by combining DECT parameters and morphologic features.

Project description:BackgroundEarly identification of the malignant propensity of pulmonary ground-glass nodules (GGNs) can relieve the pressure from tracking lesions and personalized treatment adaptation. The purpose of this study was to develop a deep learning-based method using sequential computed tomography (CT) imaging for diagnosing pulmonary GGNs.MethodsThis diagnostic study retrospectively enrolled 762 patients with GGNs from West China Hospital of Sichuan University between July 2009 and March 2019. All patients underwent surgical resection and at least two consecutive time-point CT scans. We developed a deep learning-based method to identify GGNs using sequential CT imaging on a training set consisting of 1524 CT sections from 508 patients and then evaluated 256 patients in the testing set. Afterwards, an observer study was conducted to compare the diagnostic performance between the deep learning model and two trained radiologists in the testing set. We further performed stratified analysis to further relieve the impact of histological types, nodule size, time interval between two CTs, and the component of GGNs. Receiver operating characteristic (ROC) analysis was used to assess the performance of all models.ResultsThe deep learning model that used integrated DL-features from initial and follow-up CT images yielded the best diagnostic performance, with an area under the curve of 0.841. The observer study showed that the accuracies for the deep learning model, junior radiologist, and senior radiologist were 77.17%, 66.89%, and 77.03%, respectively. Stratified analyses showed that the deep learning model and radiologists exhibited higher performance in the subgroup of nodule sizes larger than 10 mm. With a longer time interval between two CTs, the deep learning model yielded higher diagnostic accuracy, but no general rules were yielded for radiologists. Different densities of components did not affect the performance of the deep learning model. In contrast, the radiologists were affected by the nodule component.ConclusionsDeep learning can achieve diagnostic performance on par with or better than radiologists in identifying pulmonary GGNs.

Dataset Information

Development of an AI system for accurately diagnose hepatocellular carcinoma from computed tomography imaging data.

Background and aims

Methods

Results

Conclusions

Publications

Development of an AI system for accurately diagnose hepatocellular carcinoma from computed tomography imaging data.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets