Muscle ultrasound classification

Ultrasound is a non-invasive imaging technique that is increasingly used to diagnose patients with suspected neuro-muscular disorders through imaging of superficial structures. Recent studies have shown benefits of neuro-muscular ultrasound for screening and diagnostic purposes. The clinical neurophysiology department of the Radboudumc uses ultrasound of the muscles for the assessment of neuro-muscular conditions. The ultrasound probe is placed at specific locations on the body to visualize the superficial muscular tissue. At each location the probe is used to create three images by repeatedly placing the probe in the same location.

Currently, a manual ROI is drawn in each image by a technician and the mean pixel value in the ROI averaged over the three images is used to obtain a 'Z-score'. The Z-score is compared to a curve that shows the average Z-score of a healthy subject for a given age. If the Z-score is two standard deviations or more above the curve it is considered abnormal. The creation of such curves is time-consuming and expensive. Automatic analysis of the entire images could possibly be used to distinguish between abnormal and healthy subjects, negating the use of manual interaction and standard curves.

The RTC Deep Learning has developed an algorithm for the binary classification of ultrasound images of the tibialis anterior muscle. We are looking at the possibility of further extending this to classifying a combination of ultrasound images obtained from different muscle groups of the same individual.

People

Juerd Wijntjes

Juerd Wijntjes

Neurologist

Neurology, Radboudumc

Jonne Doorduin

Jonne Doorduin

Assistant professor

Neurology, Radboudumc

Ajay Patel

Ajay Patel

Coordinator RTC Deep Learning

Bram van Ginneken

Bram van Ginneken

Professor

Diagnostic Image Analysis Group