Table 1 List of identified diagnosis/classification studies along with their main characteristics
From: Deep learning-enhanced nuclear medicine SPECT imaging applied to cardiac studies
No. | First author | Year | Title | Input | Learning algorithm | Reference | Outcome | Validation (total num. of images) | Explainability | Results |
|---|---|---|---|---|---|---|---|---|---|---|
|  | Kenichi Nakajima | 2017 | Diagnostic accuracy of an artificial neural network compared with statistical quantitation of myocardial perfusion images: a Japanese multicenter study | SPECT | ANN | Human reader | Normal or abnormal | Hold out: 1001 training, 364 test (the test set is multicenter) (total num images: 1001) | Indirectly via segmenting the images into sub-images | AUC: 0.89–0.92 for stenosis > = 50% AUC: 0.58–0.72 for stenosis > = 75% |
|  | Julian Betancur | 2018 | Deep learning for prediction of obstructive disease from fast myocardial perfusion SPECT: a multicenter study | SPECT | CNN (hand-crafted) | ICA | CAD\No-CAD and per-vessel | 10FCV (1,638) | – | Per-patient sensitivity improved from 0.79 (TPD) to 0.82 (DL) Per-vessel sensitivity improved from 0.64 (TPD) to 0.69 (DL) |
|  | Julian Betancur | 2018 | Prognostic value of combined clinical and myocardial perfusion imaging data using machine learning | SPECT + age, sex, and risk factors | LogitBoost | ICA | MACE risk | Stratified 10FCV (Total num images: 2619) |  | AUC: 0.8 |
| Â | Nathalia Spier | 2019 | Classification of polar maps from cardiac perfusion imaging with graph-convolutional neural networks | Polar Maps | Graph CNNs (hand-crafted) | Human reader | Normal and abnormal | Hold-out for localization and 4FCV for the entire dataset (946 subjects) | Indirectly via segmenting the images into sub-images | Agreement rating (segment-by-segment): 0.83 Sensitivity: 0.47 Specificity: 0.70 |
|  | Julian Betancur | 2019 | Deep learning analysis of upright-supine high-efficiency SPECT myocardial perfusion imaging for prediction of obstructive coronary artery disease: a multicenter study | SPECT | CNN (hand-crafted) | ICA | CAD\No-CAD and per-vessel risk | Leave-One-Center-Out CV (4 groups, 1160 images) | – | Per-patient: AUC of 0.81 (DL) versus AUC of 0.78 (cTPD) Per-vessel: AUC of 0.77 (DL) versus AUC of 0.729 AUC (cTPD) |
|  | Rahmani | 2019 | Improved diagnostic accuracy for myocardial perfusion imaging using artificial neural networks on different input variables including clinical and quantification data | Polar maps + cardiac risk factors | ANN | ICA | CAD\No-CAD | 14 cases for validation (total num mages: 93) |  | Accuracy: 0.857 |
|  | Selcan Kaplan Berkaya | 2020 | Classification models for SPECT myocardial perfusion imaging | SPECT | CNN (VGG-19) | Human reader | Normal or abnormal | Training (66%), validation (17%), and test (17%) (Total num images: 192) | – | Accuracy: 0.93 Sensitivity: 1.0 Specificity: 0.86 |
| Â | Yuka Otaki | 2020 | Diagnostic accuracy of deep learning for myocardial perfusion imaging in men and women with a high-efficiency parallel-hole-collimated cadmium-zinc-telluride camera: multicenter study | Polar Maps | CNN | ICA | CAD\No-CAD | Leave-one-center-out external validation (4 centers in total) (Total num images: 1160) | Grad-CAM | Sensitivity: DL (0.82), SSS (0.75), U-TPD (0.77) and S-TPD (0.73) in men DL (0.71), SSS (0.71), U-TPD(0.7) and S-TPD(0.65) in women |
|  | Apostolopoulos | 2020 | Automatic characterization of myocardial perfusion imaging polar maps employing deep learning and data augmentation | Polar Maps | CNN (VGG-16) | ICA | CAD\No-CAD | 10FCV (216 subjects) | – | DL: Accuracy of 0.74, Sensitivity of 0.75. Specificity of 0.73. Similar results with the experts Semi-Quantitative Analysis: Accuracy of 0.66 |
|  | Apostolopoulos | 2020 | Multi-input deep learning approach for cardiovascular disease diagnosis using myocardial perfusion imaging and clinical data | Polar Maps + Clinical | CNN (Inception V3) + Random Forest | ICA | CAD\No-CAD | 10FCV (566 subjects) | – | Expert Accuracy: 0.7 Sensitivity: 0.89 Specificity: 0.71 Model Accuracy: 0.78 Sensitivity: 0.77 Specificity: 0.79 Cohen's Kappa: 72.24 |
| Â | Lien-Hsin Hu | 2020 | Machine learning predicts per-vessel early coronary revascularization after fast myocardial perfusion SPECT: results from multicentre REFINE SPECT registry | 18 clinical variables, 9 stress-test variables, and 28 imaging variables | LogitBoost | ICA | CAD\No-CAD Per-vessel and per-patient | 10FCV (Total num images: 1980) | Â | Per-vessel AUC: 0.79 Per-patient per-vessel AUC: 0.81 |
|  | Baskaran | 2020 | Machine learning insight into the role of imaging and clinical variables for the prediction of obstructive coronary artery disease and revascularization: an exploratory analysis of the CONSERVE study | SPECT + BMI, age, and angina severity | XGBoost | ICA | CAD\No-CAD | 5FCV (total num images: 719) |  | AUC: 0.779 |
| Â | Trung | 2020 | A deep learning method for diagnosing coronary artery disease using SPECT images of heart | SPECT | VGG-16 | Human reader | CAD\No-CAD | 5FCV (total num images: 1413) | Â | Precision: 0.823 |
| Â | Jui-Jen Chen | 2021 | Convolutional neural network in the evaluation of myocardial ischemia from CZT SPECT myocardial perfusion imaging: comparison to automated quantification | Gray SPECT images | 3D-CNN (hand-crafted) | Not disclosed | CAD\No-CAD | 5FCV (979 subjects) | Grad-CAM | Accuracy: 0.87 Sensitivity: 0.81 Specificity: 0.92 |
|  | Hui Liu | 2021 | Diagnostic accuracy of stress-only myocardial perfusion SPECT improved by deep learning | Stress only SPECT | Resnet-34 | Human reader | Normal or Abnormal | 5FCV (total num images: 37,243) | – | AUC: 0.872 ± 0.002 |
|  | Papandrianos | 2021 | Automatic diagnosis of coronary artery disease in SPECT myocardial perfusion imaging employing deep learning | SPECT | RGB-CNN (hand-crafted) | Human reader | Normal or Abnormal | Hold out 85–15% (244 subjects) | – | Accuracy: 0.93 ± 0.28 AUC: 0.936 |
| Â | de Souza Filho | 2021 | Machine learning algorithms to distinguish myocardial perfusion SPECT polar maps | SPECT | Random forest | Human reader | Normal or Abnormal | 10FCV (total num mages: 1007) | Â | AUC: 0.853 Accuracy: 0.938 Precision: 0.968 Sensitivity: 0.963 |
|  | Arvidsson | 2021 | Prediction of obstructive coronary artery disease from myocardial perfusion scintigraphy using deep neural networks | Polar maps + angina symptoms and age | CNN | ICA | Probability of CAD in the left anterior artery left circumflex artery and right coronary artery | 5FCV (total num images: 760) |  | Per-vessel AUC: 0.89 Per-patient AUC: 0.95 |
| Â | Zahiri | 2021 | Deep learning analysis of polar maps from SPECT myocardial perfusion imaging for prediction of coronary artery disease | Polar maps | CNN | Human reader | Normal or abnormal | 5FCV (total num images: 3318) | Â | Accuracy:0.7562 Sensitivity:0.7856 Specificity: 0.7434 F1 score: 0.6646 AUC: 0.8450 |
|  | Papandrianos | 2022 | Exploring classification of SPECT MPI images applying convolutional neural networks | SPECT | CNN (hand-crafted) | Human reader | Normal or abnormal | 87 cases for validation (Total num images: 314) | – | Accuracy: 0.90 AUC: 0.937 |
| Â | Papandrianos | 2022 | Deep learning exploration for SPECT MPI polar map images classification in coronary artery disease | Polar maps | RGB-CNN | Human reader | Normal and abnormal | 5FCV (Total num images: 314) | Â | Accuracy: 0.92 |
| Â | Miller | 2022 | Explainable deep learning improves physician interpretation of myocardial perfusion imaging | SPECT | DL model | ICA | CAD\No-CAD | (Total num images: 240) | Explainability | When the readers use DL AUC: 0.779 When the readers do not use DL AUC: 0.747 When DL runs autonomously AUC: 0.793 |
| Â | Ananya Singh | 2022 | Direct risk assessment from myocardial perfusion imaging using explainable deep learning | Stress rest polar maps combined with age, sex, and cardiac volumes | CNN (hand-crafted) | ICA | Prediction of death or nonfatal myocardial infarction (MI) | 10FCV (20,401 subjects) | Grad-CAM | AUC: 0.76 (internal) AUC: 0.73 (external) |