Research on Automatic Classification of Distal Radius Fractures Based on Deep Learning

Feng Yang; Rikun Cong; Weiguo Wang; Bo Ding

doi:10.3788/LOP202158.1210020

Journals >Laser & Optoelectronics Progress >Volume 58 >Issue 12 >Page 1210020 > Article

Laser & Optoelectronics Progress
Vol. 58, Issue 12, 1210020 (2021)

Research on Automatic Classification of Distal Radius Fractures Based on Deep Learning

Feng Yang¹, Rikun Cong¹, Weiguo Wang², and Bo Ding^1、*

Author Affiliations

¹Network Information Center of Shandong University of Traditional Chinese Medicine, Jinan, Shandong 250355, China

²The First Clinical College of Shandong University of Traditional Chinese Medicine, Jinan, Shandong 250355, China

show less

DOI: 10.3788/LOP202158.1210020 Cite this Article Set citation alerts

Feng Yang, Rikun Cong, Weiguo Wang, Bo Ding. Research on Automatic Classification of Distal Radius Fractures Based on Deep Learning[J]. Laser & Optoelectronics Progress, 2021, 58(12): 1210020 Copy Citation Text

show less

$Examples of 3 specific fracture types of DRF. (a) DRF-A; (b) DRF-B; (c) DRF-C$

Fig. 1. Examples of 3 specific fracture types of DRF. (a) DRF-A; (b) DRF-B; (c) DRF-C

Download full size

Fig. 2. DRF images before and after processing by CLAHE algorithm. (a) DRF-A; (b) DRF-B; (c) DRF-C

Download full size

Fig. 3. Main structure of DRF-Net model

Download full size

Fig. 4. DRF-Net training process. (a) Schematic of accuracy rate and iteration number; (b) schematic of loss value and iteration number

Download full size

Feature extraction algorithm	Feature vector dimension
Gabor feature	48
LBP feature	59
GLCM feature	8
HOG feature	324
Gray feature	6

Table 1. Five traditional feature extraction algorithms and their specific dimensions

Feature extraction algorithm	Classifier	Validation set A_Acc%	Test set A_Acc%	F₁	AUC
Gabor feature		67.8±1.3	68.4±2.4	0.737	0.664
LBP feature		58.7±2.7	57.7±1.1	0.621	0.552
GLCM feature	SVM	69.4±2.2	72.5±1.3	0.731	0.683
HOG feature		65.1±2.6	69.4±2.1	0.714	0.626
Gray feature		60.1±3.2	62.2±3.4	0.628	0.542
Gabor feature		68.5±2.7	72.1±1.2	0.708	0.711
LBP feature		68.1±3.1	70.8±1.2	0.721	0.674
GLCM feature	ELM	62.5±1.4	67.1±2.7	0.692	0.670
HOG feature		67.2±1.3	65.4±1.4	0.734	0.702
Gray feature		54.2±1.9	52.9±2.4	0.533	0.556
Feature extraction algorithm	Classifier	Validation set A_Acc%	Test set A_Acc%	F₁	AUC
Gabor feature		72.1±2.3	70.1±1.2	0.724	0.679
LBP feature		65.5±1.2	64.2±1.9	0.679	0.604
GLCM feature	RF	68.7±1.9	69.4±2.1	0.739	0.626
HOG feature		63.9±1.6	67.8±1.4	0.704	0.725
Gray feature		57.6±2.4	59.3±2.8	0.604	0.571
Gabor feature		67.3±1.5	68.8±3.2	0.697	0.592
LBP feature		71.4±2.8	65.9±4.1	0.721	0.661
GLCM feature	kNN	63.5±2.1	59.4±4.1	0.693	0.685
HOG feature		65.9±2.6	58.8±4.4	0.701	0.729
Gray feature		51.6±3.6	52.3±2.2	0.671	0.507
DRF-Net	Softmax	78.3±3.7	79.1±2.4	0.815	0.792
DRF-Net+TL	Softmax	84.2±2.3	82.1±0.9	0.834	0.826

Table 2. Classification performance statistics table of different traditional classification methods

Deep learning model	Validation set A_Acc%	Test set A_Acc%	F₁	AUC
AlexNet	76.7±1.8	75.6±2.2	0.732	0.774
GoogleNet	84.5±1.6	80.2±2.1	0.793	0.812
ResNet50	83.5±3.7	80.2±2.8	0.822	0.806
DRF-Net	79.3±2.1	76.9±1.7	0.815	0.792
DRF-Net+TL	84.2±2.3	83.1±0.9	0.834	0.826

Table 3. Comparison of classification performance between DRF-Net and other deep learning models

Feng Yang, Rikun Cong, Weiguo Wang, Bo Ding. Research on Automatic Classification of Distal Radius Fractures Based on Deep Learning[J]. Laser & Optoelectronics Progress, 2021, 58(12): 1210020

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information