Table 8.

Summary of 3D CNN classification approaches.

Publication reference	Target tasks	Modality (imaging sequence)	Data set	Network architecture	Performance
Tolpadi et al. [21]	Predict total knee replacement	MRI (3D-DESS)	OAI: 4790 subjects (3114 training, 957 validation, 719 testing)	DenseNet-121	AUC ± SD: 0.886 ± 0.020
Pedoia et al. [20]	Detect and stage severity of meniscus and patellofemoral cartilage lesions	MRI (3D-FSE CUBE)	1478 images (training : validation : testing: 65 : 20 : 15%)	3D CNN	AUC ± SD: 0.89 (menisci), 0.88 (cartilage); SN: 89.81% (menisci), 80.0% (cartilage); SP: 81.98% (menisci), 80.27% (cartilage)
Nunes et al. [19]	Stage severity of cartilage lesion	MRI (3D-FSE CUBE)	1435 images (training : validation : testing: 65 : 20 : 15%)	3D CNN	Accuracy: 86.7%
Zhang et al. [72]	Detect anterior cruciate ligament lesion	MRI (PDW-SPAIR)	(285 training, 81 validation, 42 testing) images	3D DenseNet	AUC: 0.960; accuracy: 0.957; SN: 0.944; SP: 0.940

Note. Modality (imaging sequence): magnetic resonance imaging (MRI); data set: Osteoarthritis Initiative (OAI); network architecture: convolutional neural network (CNN); performance: specificity (SP), sensitivity (SN), area under receiver operating characteristics curve (AUC), and standard deviation (SD).