Skip to main content

Table 1 Data augmentation strategy and details

From: Machine learning-based prediction of the necessity for the surgical treatment of distal radius fractures

Augmentation Type

Function

Parameter

Augmentation Details

Probability

Cropping

Random Resized Crop

Crop Area Scale

A value is randomly selected between -50% and 50% of the original image area is cropped and resized into output size (output size = 512 × 512 × 3)

1.0

Flip

Horizontal Flip

Flip Direction

The image is flipped horizontally (mirrored along the vertical axis)

0.5

Color Adjustment

Color Jitter

Brightness

A value is randomly selected between -50% and 50% of the original value

0.5

Contrast

A value is randomly selected between -50% and 50% of the original value

0.5

Saturation

A value is randomly selected between -50% and 50% of the original value

0.5

Hue

A value is randomly selected between -10% and 10% of the original value

0.5

Solarization

Threshold

Any pixel with a value above 128 of the images (on a scale of 0–255) are inverted (original value range = 0–255)

0.5

Brightness/Contrast

Random Brightness Contrast

Brightness Limit

A value is randomly selected between -20% and 20% of the original value

0.2

Optical Distortion

Optical Distortion

Distort Limit

A value is randomly selected between -5% and 5% of the original value

0.5

Shift Limit

A value is randomly selected between -5% and 5% of the original value