Automatic and standardized quality assurance of digital mammography and tomosynthesis with deep convolutional neural networks

Insights into Imaging

Table 1 Results of models training and validation

Type	MODEL	Projection	Training images	Validation images	Training accuracy /training RSME [mm]	Validation accuracy /validation RSME [mm]
Classification	Parenchyma	CC	4750	1357	89.4%	87.2%
	Parenchyma	MLO	3861	1103	88.5%	86.9%
	IMF	MLO	2826	807	84.6%	84.9%
	Pectoralis	CC	1255	359	88.1%	92.7%
	Nipple	CC + MLO	1155	330	87.3%	86.9%
Regression	Nipple	CC + MLO	2720	777	8.3	8.4
	Pectoralis cranial	MLO	1467	419	8.2	12.9
	Pectoralis caudal	MLO	1467	419	8.3	8.4
	Pectoralis angle	MLO	1467	419	n/a	n/a
	Posterior-Nipple line	MLO	1467	419	10.5	11.3

Results for Pectoralis Angle and Posterior-Nipple Line have been calculated based on Nipple and Pectoralis positions