Deep learning in hepatocellular carcinoma: Current status and future perspectives

Table 1 Studies applying deep learning for hepatocellular carcinoma

Study	Cohort	Data source	Deep learning	Input	Output	Main findings
Predicting HCC risk using clinical variables
Ioannou et al[14] 2020	48151 HCV cirrhosis (T: 90%, V: 10%)	VHA database	RNN	Clinical variables	Risk of HCC development	RNN predicted HCC development with AUC of 0.759, and AUC of 0.806 among those who achieved SVR
Phan et al[15] 2020	6052 HBV and HCV (T: 70%, V: 30%)	Taiwanese NHIRD	CNN	Disease history data	Risk of HCC development	CNN achieved an accuracy of 0.980 and AUC of 0.886 for predicting HCC development among viral hepatitis patients
Nam et al[16] 2020	T: 424 HBV cirrhosis; V: 316 HBV cirrhosis	2 Korean centers	ResNet	Clinical variables	Risk of HCC development	DL model achieved an accuracy of 0.763 and AUC of 0.782 in the validation cohort and outperformed previous models
Nam et al[17] 2020	T: 349 LT recipients; V: 214 LT recipients	3 Korean LT centers	ResNet	Clinical variables	Recurrent HCC after LT	DL model significantly outperformed conventional models in prediction of post-T HCC recurrence with AUC of 0.75
Multi-omics-based HCC diagnosis and prognostication
Xie et al[20] 2018	T: 133 HCC/54 HV; V: 52 HCC/34 HV	1 center in China	ANN	Gene expression	HCC detection	ANN using nine genes had an AUC of 0.943, 98% sensitivity, and 85% specificity for classifying HCC
Choi et al[21] 2018	135 HCC (10-fold CV)	TCGA	G2Vec	Gene expression	HCC prognosis	G2Vec showed significantly higher prediction accuracy for patient outcomes compared to existing gene selection tools
Chaudhary et al[22] 2018	T: 360 HCC; V: 220, 221, 166, 40, 27 HCC	TCGA; 5 external datasets	Auto-encoder	RNA-seq, miRNA-seq, methylation	HCC prognosis	DL model distinguished groups with survival differences and identified mutations and pathways predicting aggressive tumor behavior
Radiology-based HCC diagnosis/prediction
Streba et al[25] 2012	112 FLL (10-fold CV)	1 center in Romania	ANN	US images	FLL type	ANN had 87.12% testing accuracy, 93.2% sensitivity, and 89.7% specificity for classifying 5 classes of liver lesions
Hassan et al[26] 2017	110 FLL (10-fold CV)	1 center in Egypt	Auto-encoder	US images	FLL type	The proposed system had 97.2% accuracy, 98% sensitivity, and 95.70% specificity for classifying liver lesions
Bharti et al[27] 2018	24 normal, 25 CLD, 25 cirrhosis, 20 HCC	1 center in India	CNN	US images	Liver stages	CNN achieved 96.6% classification accuracy for differentiating normal liver, CLD, cirrhosis, and HCC
Schmauch et al[28] 2019	T: 367 FLL; V: 177 FLL	Centers in France	ResNet	US images	FLL type	DL model reached mean AUC of 0.935 for focal liver lesion detection and 0.916 for focal liver lesion characterization
Brehar et al[29] 2020	T: 200 HCC; V: 68 HCC	1 center in Romania	CNN	US images	HCC detection	CNN achieved AUC of 0.95, accuracy of 0.91, 94.4% sensitivity and 88.4% specificity for HCC detection
Jin et al[30] 2021	434 HBV (3:1:1 split)	1 center in China	DL radiomics	US images	Risk of HCC development	DL radiomics model predicted 5-yr HCC development risk with AUC of 0.900 in the test set
Yasaka et al[31] 2018	T: 460 liver masses; V: 100 liver masses	1 center in Japan	CNN	CT images	Liver mass type	CNN classified liver lesions into five categories with a median AUC of 0.92
Shi et al[32] 2020	449 FLL; (T: 80%, V: 20%)	1 center in China	CNN	CT images	FLL type	CNN applied to three-phase CT protocol images achieved AUC of 0.925 for differentiating HCC from other FLLs
Hamm et al[43] 2019	T: 434 FLL; V: 60 FLL	1 center in United States	CNN	MRI images	FLL type	CNN achieved 90% sensitivity and 98% specificity for classifying FLLs and AUC of 0.992 for HCC classification
Wang et al[44] 2019	T: 434 FLL; V: 60 FLL	1 center in United States	CNN	MRI images	FLL type	Interpretable DL system achieved 76.5% PPV and 82.9% sensitivity for identifying correct radiological features
Wu et al[45] 2020	89 liver tumors; (60: 20: 20)	1 center in United States	CNN	MRI images	LI-RADS grading	CNN achieved AUC of 0.95, 90% accuracy, 100% sensitivity and 83.5% PPV for LI-RADS grading of liver tumors
Zhen et al[46] 2020	T: 1210 liver tumors; V: 201 liver tumors	1 center in China	CNN	MRI images	Liver tumor type	CNN combined with clinical data showed AUC of 0.985 for classifying HCC with 91.9% agreement with pathology
Radiology-based HCC prognostication, treatment planning, and response to treatment
Zhang et al[47] 2021	T: 158 HCC; V: 79 HCC	1 center in China	CNN	MRI images	MVI in HCC	CNN achieved AUC of 0.72, 55% sensitivity, and 81% specificity for preoperative MVI in HCC patients
Wang et al[48] 2020	T: 60 HCC; V: 40 HCC	1 center in China	CNN	MRI images	MVI in HCC	Fusion of deep features from MRI images yielded AUC of 0.79 for MVI prediction in HCC patients
Jiang et al[49] 2021	405 HCC; (T: 80%, V: 20%)	1 center in China	CNN	CT images	MVI in HCC	CNN achieved AUC of 0.906 for prediction of MVI. Mean survival was significantly better in the group without MVI
An et al[50] 2020	141 single HCC resect MWA	1 center in China	CNN	MRI images	Ablative margin	Deep learning model accurately estimated ablative margins and risk of local tumor progression
Liu et al[51] 2020	T: 89 HCC resect TACE; V: 41 HCC rec. TACE	1 center in China	CNN	Ultrasound images	Response to TACE	Deep learning radiomics model predicted tumor response to TACE with AUC of 0.93
Peng et al[52] 2020	T: 562 HCC resect TACE; V:227 HCC rec. TACE	3 centers in China	CNN	CT images	Response to TACE	Deep learning model had accuracies of 85.1% and 82.8% for predicting TACE response in 2 validation cohorts
Liu et al[53] 2020	243 HCC resect TACE (6:1:3 split)	1 center in China	CNN	CT images	Post-TACE survival	Higher DL score was an independent prognostic factor and predicted overall survival with AUCs of 0.85-0.90
Zhang et al[54] 2020	201 HCC resect TACE + sorafenib (T: 120, V: 81)	3 centers in China	CNN	CT images	OS on TACE + sorafenib	Deep learning signature achieved C-index of 0.714 for predicting OS in HCC patients receiving TACE + sorafenib
Histopathology-based HCC diagnosis, subtyping, and outcome predictions
Lin et al[55]2019	113 HCC	1 center in China	CNN	Histopath images	HCC differentiation	CNN achieved an accuracy of 0.941 for determining HCC differentiation on multiphoton microscopy
Kiani et al[56] 2020	70 WSI (35 HCC, 35 CC)	TCGA	CNN	Histopath images	HCC vs CC	CNN-based “Liver Cancer Assistant” accurately differentiated HCC vs cholangiocarcinoma
Liao et al[57] 2020	T: 491 HCC; V: 455 HCC	TCGA; 1 center in China	CNN	Histopath images	HCC detection, mutations	CNN distinguished HCC from adjacent tissues with AUC of 1.00 and predicted specific mutations with AUC over 0.70
Wang et al[58] 2020	T: 99 HCC; V: 205 HCC	TCGA	CNN	Histopath images	Histological HCC subtype	Unsupervised clustering identified 3 histological subtypes complementing molecular pathways and prognostic value
Chen et al[59] 2020	T: 402 HCC/89 normal; V: 67 HCC/34 normal	GDC portal; 1 center in China	CNN	Histopath images	HCC grade mutations	CNN achieved 89.6% accuracy for tumor differentiation stage and predicted presence of specific gene mutations
Lu et al[60] 2020	421 HCC/105 normal (6-fold CV)	GDC portal	CNN	Histopath images	HCC prognosis	Pre-trained CNN predicted OS using pathology images and identified HCC subgroups with different prognosis
Saillard et al[61] 2020	T: 194 HCC; V: 328 HCC	1 French center TCGA	CNN	Histopath images	Survival after HCC resection	CNN models using pathology images predicted survival with C-index 0.75-0.78 and outperformed conventional models
Shi et al[62] 2021	T: 1125 HCC; V: 320 HCC	1 center in China; TCGA	CNN	Histopath images	HCC outcomes	Deep learning-based “tumor risk score” was superior to clinical staging and stratified 5 groups of different prognosis
Yamashita et al[63] 2021	T: 36 WSI; V: 30 WSI	1 center in United States; TCGA	CNN	Histopath images	Post-surgical recurrence	CNN risk scores outperformed TNM system for predicting recurrence and identified high-and low-risk subgroups

ANN: Artificial neural network; AUC: Area under the curve; CC: Cholangiocarcinoma; CNN: Convolutional neural network; CV: Cross-validation; FLL: Focal liver lesion; GDC: Genomic Data Commons; HBV: Hepatitis B virus; HCC: Hepatocellular carcinoma; HCV: Hepatitis C virus; HV: Healthy volunteers; LT: Liver transplant; MVI: Microvascular invasion; MWA: Microwave ablation; OS: Overall survival; PFS: Progression-free survival; RFA: Radio-frequency ablation; RNN: Recurrent neural network; SR: Surgical resection; STS-net: Spatial transformed similarity network; SVR: Sustained virologic response; T: Training; TCGA: The Cancer Genome Atlas; V: Validation; VHA: Veterans Health Administration; WSI: Whole slide image; CT: Computed tomography; MRI: Magnetic resonance imaging; NHIRD: National Health Insurance Research Database; TNM: Tumor, Nodes, Metastasis; TACE: Trans-arterial chemoembolization; LI-RADS: Liver Imaging Reporting and Data System.