Skip to main content

Table 2 Results of model accuracies and errors on test datasets

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

Type

MODEL

Projection

Test images

Test accuracy/test RSME [mm]

Sensitivity TPR

Specificity TNR

Precision PPV

Classification

Parenchyma

CC

1944

97.1% [95% CI 96.2–97.7%]

99.8% [95% CI 99.3–99.9%]

95.4% [95% CI 93.7–96.3%]

93.4% [95% CI 91.6–94.8%]

Parenchyma

MLO

87.3% [95% CI 86.3–88.2%]

89.0% [95% CI 87.4–90.2%]

87.6% [95% CI 85.0–89.2%]

93.6% [95% CI 91.7–94.8%]

IMF

MLO

973

96.3% [95% CI 94.9–97.4%]

99.3% [95% CI 98.2–99.7%]

93.1% [95% CI 90.4– 95.2%]

93.6% [95% CI 91.4–95.3%]

Pectoralis

CC

971

98.5% [95% CI 98.2–99.5%]

99.5% [95% CI 98.6–99.9%]

96.3% [95% CI 93.7–98.1%]

98.1% [95% CI 96.8–98.9%]

Nipple

CC + MLO

1944

93.0% [95% CI 91.7–94.0%]

94.0% [95% CI 92.8–95.1%]

69.2% [95% CI 57.9–78.9%]

98.5% [95% CI 98.0–98.9%]

Regression

Nipple

CC + MLO

350

8.4 [95% CI 7.3–9.5]

n/a

Pectoralis Cranial

MLO

188

17.9 [95% CI 13.2–21.0]

n/a

Pectoralis Caudal

MLO

188

13.2 [95% CI 9.9–16.0]

n/a

Pectoralis Angle

MLO

973

99.1% [95% CI 98.2–99.5%]

99.0% [95% CI 98.1–99.6%]

99.0% [95% CI 96.6–99.8%]

99.7% [95% CI 98.9–99.9%]

Posterior-Nipple Line

MLO

973

99.0% [95% CI 98.1–99.1%]

99.7% [95% CI 98.9–99.9%]

96.9% [95% CI 94.1–98.7%]

98.8% [95% CI 97.8–99.4%]