Table 2.
Overview of ML algorithms applied to hepatocellular carcinoma diagnosis.
| Authors | ML Algorithm | Aim | Imaging Modality | Performance |
|---|---|---|---|---|
| Bharti et al. [33] | CNN based on ensemble model (k-NN, SVM and rotation forest) | Classify four classes of liver images on US, namely normal liver, chronic liver disease, cirrhosis and HCC | US | Accuracy: 96.6% |
| Schmauch et al. [34] | ResNet50 Neural Network | Detect and classify liver lesions as benign or ultrasound malignant | US | AUC: 0.93 and 0.91 * |
| Hassan et al. [35] | stacked sparse auto-encoder with SoftMax layer classifier | Detect HCC, hemangioma and liver cysts | US | Sensitivity: 98% Specificity: 95.7% |
| Guo et al. [36] | deep canonical correlation analysis-multiple kernel learning based classifier | Discriminate benign and malignant liver lesions | CEUS | Accuracy: 90.41 ± 5.80% |
| Mokrane et al. [37] | k-NN, SVM and RF | Classify hepatic nodules as HCC or non-HCC | CT | AUC: 0.66 |
| Yasaka et al. [38] | CNN | Classification of liver lesions in five categories | CT | AUC: 0.92 |
| Raman et al. [39] | RF | Classification of hypervascular liver lesions | CT | Accuracy: 90% |
| Nayak et al. [32] | SVM | Diagnosis of cirrhosis and hepatocellular carcinoma | CT | DICE score: 90%, 86% and 81% ** |
| Vivanti et al. [40] | CNN | Detection of tumor recurrence based on CT volume/tumor load | CT | Accuracy: 86% |
| Wenqi et al. [41] | CNN | Diagnostic accuracy of a three-phase CT protocol without PV vs. four-phase CT protocol | CT | Accuracy: 85.6% vs. 83.3% |
| Yamada et al. [42] | CNN | Diagnosis of primary liver cancers using transfer learning | CT | Mean DP: 44.1%, 44.2%, and 43.7% *** |
| Hamm et al. [43] | CNN | Classify liver lesions | MRI | Accuracy: 92% |
| Wu et al. [44] | CNN | CNN model for LI-RADS grading | MRI | AUC: 0,95 **** |
| Jansen et al. [45] | extremely randomized trees classifier | Automated classification system cataloguing liver lesions as adenoma, cyst, hemangioma, HCC and metastasis | MRI | Sensitivity/Specificity: 80/78%, 93/93%, 84/82%, 73/56% and 62/77% ***** |
| Zhen et al. [46] | CNN | Detecting and categorizing liver tumors | MRI | AUC: 0.98, 0.99, 0.96 ****** |
| Preis et al. [47] | ANN | Analyze 18F-FDG PET-CT liver uptake of patient at risk of developing HCC | PET | AUC: 0.89 |
ML: machine learning; CNN: convolutional neural network; k-NN: k-nearest neighbor; SVM: support vector machine; RF: random forest; DL: deep learning; PV: portal venous; AUC: area under the curve; DP: diagnostic performance; ANN: artificial neural network. * Results respectively for focal liver lesion detection and focal liver lesion characterization. ** Results respectively for healthy liver, cirrhosis and HCC. *** Results respectively by pixel shifts, rotations, and skew misalignments transfer learning methods. **** Result for differentiation between LR-3 and LR-4/LR-5 tumors. ***** Results respectively for adenoma, cyst, hemangioma, HCC and metastasis detection. ****** Results respectively for HCC, metastasis and other primary malignancies.