Bioinformatics Seminars

Current Bioinformatics Seminar

22 October 2019

Deep learning model to estimate skeletal muscle loss in non-small cell lung cancer patients

Kaushalya Amarasinghe

Recently, convolutional neural networks (CNNs) have successfully been used in different image analysis tasks. In particular, CNNs have been successfully applied for biomedical image segmentation. Skeletal muscle index is one of the validated methods to recognise muscle wastage (sarcopenia) due to complications in cancer and cancer therapies. However, calculation of SMI currently requires multiple manual steps including segmenting the muscle in CT scans using expensive proprietary software, expert knowledge and experience.

In this study, we trained a CNN based model to automatically delineate skeletal muscle at 3rd lumbar (L3) vertebra of CT scans. We trained the model using scans from a cohort of 66 patients. The model was tested on a separate independent cohort of 43 patients. I will present how the model compares against the manual segmentation and how it performs in classifying patients with sarcopenia.

Search past seminars