Research group | MRI sequence | AI method for classification | Dataset | AI task | Acc (%) | AUC |
---|---|---|---|---|---|---|
Zhang et al. [23] | Nonenhanced cine | Fully connected discriminative network (DL) | 299 patients (MI: nā=ā212) | Detecting and delineating chronic MI. Classification as normal or infarcted myocardium. | - | 0.94 |
Snaauw et al. [18] | Cine | DenseNet (DL) | ACDC (DC, HCM, MI, ARV, NOR) | End-to-end diagnosis and segmentation. | 78 | - |
Khened et al. [19] | Cine | Random forest method (ML) | ACDC (DC, HCM, MI, ARV, NOR) | Fully automated segmentation and classification. | 90 | - |
Ammar et al. [8] | Cine | Classifier ensemble combining multilayer perceptron, random forest, and support vector machine (ML) | ACDC (DC, HCM, MI, ARV, NOR) | Automated pipeline for segmentation and prediction. | 92 | - |
Agibetov et al. [20] | LGE Cine T1 mapping | VGG16 CNN pretrained on ImageNet (DL) | 502 patients (CA:Ā nĀ = 82) | Detection of potential patterns of CA. | 94 | 0.96 |
Ohta et al. [12] | MDE | GoogLeNet AlexNet ResNet-152 CNNs (DL) | 200 patients | Detection and classification of MDE patterns. | 78.9 to 82.1 | 0.938Ā to 0.948 |
Martini et al. [21] | LGE | 3 pretrained CNNs (DL), Comparison to gradient boosting classifier (ML) | 206 patients with suspected CA | Automated classification as amyloidosis present or absent based on average probability from the 3 CNNs. | 88 (DL) 90 (ML) | 0.982 (DL) 0.952 (ML) |
El-Rewaidy et al. [22] | Native T1 mapping | Linear support vector machine and regression model (ML) | 321 patients (Control, HCM, DCM) | Texture analyses on myocardial native T1 mapping to differentiate between fibrosis patterns in patients with HCM and DCM. | 89.3 | - |