First look at dataset

First look at dataset

Report of unusual cases - notebook

Goal project

• Detect anomalies in hand xray images
• Should work on unlabeled data → can't use classification task
• First step: to detect obvious anomalies like metal plates (e.g.)

Context dataset

Data is a subset of this dataset: MURA. "Study is manually labeled by ra- diologists as either normal or abnormal. To evaluate models robustly and to get an estimate of radiologist performance, we collect additional labels from six board- certified Stanford radiologists on the test set, consisting of 207 musculoskeletal studies. On this test set, the majority vote of a group of three radiologists serves as gold standard" (see abstract).

Overview over dataset

Images per person and study

The range of amount of image ranges from 1 to 5 per study. One person can have more than one study. Most of the studies consist of 2-3 images.

Distribution label

More than 1400 studies are normal. About 500 studies are not normal.

Possible problems

• The images have a frame around the hand → can influence the model

• The frame can vary

• Background color is not fixed

• Noisy

• Text labels, sometimes more than one

• Arrow in the image

• Different size

Unusual cases

1. Different contrast levels/colours

2. Not whole hand in a picture / part of other body part in the picture

3. Different odd objects

is classified as normal

4. Quality of images is different

Angles are different

Double hands

For the same view point positions are different

Only parts of a hand

Wired position

Part of other hand in the image

2 images next to another in the same scan

5. Misclassification?