From: Application and prospects of AI-based radiomics in ultrasound diagnosis
Research area | Modality | Method | Objective | Performance |
---|---|---|---|---|
Clinical application research | BUS, UE | Deep polynomial network [34] | Differentiating malignant and benign breast tumors | AUC: 0.961 |
BUS, UE | Nomogram [35] | Prediction of malignant status of breast lesions | AUC: 0.920 | |
BUS, UE | Deep learning [36] | Differentiating malignant and benign breast tumors | Accuracy: more than 90% | |
BUS, CDFI, UE | Deep learning [37] | Assessment of breast cancer risk | AUC: 0.922 for dual-modal and 0.955 for tri-modal method | |
BUS, UE | LASSO [38] | Differentiating benign, lymphomatous, and metastatic lymph nodes | AUC: 0.960 for benign vs lymphomatous, 0.716 for benign vs metastatic, 0.933 for lymphomatous vs metastatic, and 0.856 for benign vs malignant | |
BUS, UE | Scoring and support vector machine (SVM) [39] | Evaluation of axillary lymph node metastasis in breast cancer patients | AUC: 0.881 for scoring and 0.895 for SVM | |
BUS, CDFI | Deep learning [40] | Diagnosis of unexplained cervical lymphadenopathy | AUC: 0.873, 0.837, and 0.840 in the three testing cohorts for four common etiologies | |
BUS, UE | Deep learning [41] | Grading liver fibrosis | AUC: 0.950, 0.932, and 0.930 for classifying S4, ≥ S3, and ≥ S2 | |
BUS, UE and shear wave viscosity imaging | Sparse representation theory and SVM [42] | Diagnosis and clinical prediction of HCC | AUC: 0.94 for benign and malignant classification, 0.97 for malignant subtyping, 0.97 for PD-1 prediction, 0.94 for Ki-67 prediction, and 0.98 for MVI prediction | |
BUS, CDFI, UE | Deep learning [43] | Diagnosis of suspicious thyroid nodules | AUC: 0.928 | |
BUS and superb microvascular imaging ultrasound | SVM [44] | Differentiation between gallbladder neoplastic polyps and cholesterol polyps | AUC: 0.850 | |
BUS, UE, CEUS | Logistic regression and nomogram [45] | Prediction of microvascular invasion and recurrence of HCC | AUC: 0.789 | |
Algorithm research | BUS, CDFI, UE | A self-supervised multi-modal fusion network [46] | Diagnosing thyroid nodules | Accuracy: 89.79%, higher than other deep learning methods |
BUS, CDFI, shear-wave and strain UE | An modality auto-weighting and recovery framework [47] | Diagnosing breast cancer | Accuracy: 92.63% for modality completeness and 90.65% for modality missing | |
BUS, CDFI, UE and CEUS | Multi-step modality fusion network [48] | Identifying the histologic subtypes of metastatic cervical lymphadenopathy | Accuracy: 80.06%, true-positive rate: 81.81%, and true-negative rate: 80.00% | |
BUS, CDFI | Tissue-aware cervical lymph node diagnosis method via multi-modal ultrasound semantic segmentation [49] | Post-pandemic healthcare for COVID-19 vaccine | Accuracy: 82.54% | |
BUS, CEUS | Cross-modality lesion segmentation network [32] | Semantic segmentation | IOU: 80.06% for metastasis cervical lymph nodes and 75.62% for breast lesions |