Table 3 Studies using deep learning and machine learning to analyze liver biopsy images to diagnose, classify, and predict the outcome of hepatocellular carcinoma

Yamashita, Long [28]

Stanford-HCCDET;

TCGA

convolution neural network

Microscopic images of tissue

The reappearance of cancer after surgery

CNN risk scores were better than the TNM system at predicting which patients would have a recurrence of their cancer, and they were also able to identify groups of patients at high and low risk of recurrence

Saillard, Schmauch [29]

French center and

TCGA

convolution neural network

Microscopic images

Survival after HCC resection

CNNs using pathology images are a promising new tool for predicting patient survival, outperforming conventional models with a C-index of 0.75–0.78

Tohme, Yazdani [30]

TCGA-LIHC

ANN

Clinic pathological data

used individual patient tumor genomic data to develop a three-gene predictive score

ANN identified 15 genes with normalized importance > 50%

Saito, Toyoda [31]

Yamaguchi University (100 cases), Ogaki Municipal Hospital (47 cases), and Tokyo Medical University (11 cases)

SVM

pathological data

ML-based method for predicting cancer recurrence using all available information on cancer tissue

Prediction used in HCC area information (Accuracy: 88.8%)

Prediction used in non-HCC area information (Accuracy: 64.0%)

Zeng, Zeng [32]

Eastern Hepatobiliary Surgery Hospital data

Random survival forests

Clinic pathologic variables

compare the random survival forests (RSF) model with CPH models in the prediction of early recurrence for HCC patients

The time-dependent AUC (2 years) of the RSF model were 0.818 (SE = 0.008), 0.823 (SE = 0.014), and 0.785 (SE = 0.025),

[33]

TCGA

CNN

Microscopic images

classify image patches as containing either HCC or CC

Using a CNN-based 'Liver Cancer Assistant' to accurately differentiate between hepatocellular carcinoma and cholangiocarcinoma. The model had a diagnostic accuracy of 0.885

Liao, Long [34]

TCGA and a center

in China

convolution neural network

Microscopic images

HCC detection and prediction of the mutation status of HCC samples

The CNN successfully distinguished hepatocellular carcinoma from adjacent tissues with an AUC of 1.00 and accurately predicted mutations with an AUC exceeding 0.70

Wang, Jiang [35]

TCGA–LIHC

convolution neural network

Microscopic images

Cellular classification of hepatocellular carcinoma

The application of unsupervised clustering revealed the presence of three histological subtypes that complemented molecular pathways and demonstrated prognostic value. The model was accurate in the training dataset, with an overall classification of 99% for tumor cells and 97% for lymphocytes, respectively

Chen, Zhang [36]

GDC online resource center with one center in China

convolution neural network

Microscopic images

Mutations that contribute to HCC progression and metastasis

The CNN achieved an accuracy of 89.6% in predicting tumor differentiation stage and successfully predicted the presence of specific gene mutations

Lu and Daigle Jr [37]

GDC Online data repository

convolution neural network

Microscopic images

Risk of death from HCC

The pre-trained CNN utilized pathology images to predict overall survival (OS) and effectively-identified hepatocellular carcinoma (HCC) subgroups with distinct prognoses

Shi, Wang [38]

1 center in China;

TCGA

convolution neural network

Microscopic images

HCC outcomes

The deep learning-based 'tumor risk score' outperformed clinical staging and effectively stratified five groups with varying prognoses