Emergence of Deep Learning in Knee Osteoarthritis Diagnosis

Table 4

Summary of 2DCNN classification approaches on progression of osteoarthritis diagnosis.

Publication reference	Target tasks	Modality (imaging sequence)	Data set	Network architecture	Performance
Guan et al. [53]	Predict OA progression	X-ray (plain radiography)	OAI: 600 subjects (450 training, 50 validation, 100 testing)	Vgg16 and DenseNet	Vgg16: AUC: 0.717; SN: 80.0%; SP: 56.1%
					DenseNet: AUC: 0.744; SN: 94.1%; SP: 48.0%
Tiulpin et al. [54]	Predict OA progression	X-ray (plain radiography)	OAI: 5139 images (training)	CNN	AUC: 0.71
			MOST: 2,491 images (testing)
Guan et al. [11]	Predicting progression of radiographic medial joint space loss	X-ray (plain radiography)	OAI: (1400 training, 150 validation, 400 testing) images	YOLO + DenseNet	AUC: 0.799; SN: 78.0%; SP: 75.5%
Razmjoo et al. [7]	Predict OA incidence	MRI	OAI: 1805 subjects	Topological data analysis (TDA) + graph convolutional network (GCN model)	Accuracy (F1): 0.91; SN: 0.84; SP: 0.99
Li et al. [55]	Predict OA progression by assessing severity	X-ray (plain radiography)	MOST: 3021 subjects (training : validation : testing; 80 : 10 : 10%)	Siamese neural network	AUC: 0.90

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