School of EE Seminar- Mustafa Bshara-Automatic Segmentation of Rotator Cuff Tears by Models to Visual System Mechanisms
https://zoom.us/j/99639199094?pwd=MktVS3E5eUZldkxtd0p6VmY4ckYvQT09
Meeting ID: 996 3919 9094
Passcode: 0XpN1w
Electrical Engineering Systems Seminar
Speaker: Mustafa Bshara
M.Sc. student under the supervision of Dr. Hedva Spitzer
Monday, January 31st, 2022, at 15:00
Automatic Segmentation of Rotator Cuff Tears by Models to Visual System Mechanisms
Abstract
The need for segmentation rotator cuff tears (RCT), a common injury that cause pain and disability, is crucial for the type of treatment, surgical or non-surgical treatment. Automatic segmentation of ultrasonographic tendinopathy cases is challenging due to non-homogenous RCT in its shape, size and location. Only two studies have coped with this challenge by applying by artificial neural networks or semi-automatic segmentation by active contour, which includes level-set techniques. We further developed the MICO segmentation model, which is based on fuzzy c-mean algorithm and level-set framework, but we use unique features, which enable us to identify the tendon and the RCT according to three characteristic properties. These properties are multi scale texture, structure enhancement (through line completion), and referring to the spatial location. The inspiration first two features came from visual system mechanisms.
We performed a feedback process with the radiologist delineations and the algorithm segmentation and its derived corrections. The final model segmentation and radiologist feedback have been compared and verified by several metrics (accuracy: 96.56%, Dice: 87.92%, Recall: 94.71% and Precision: 92.45), while the neural network study showed Dice: 74%. Our Segmentation by Adaptive Lateral-Interactions and Multi-Scale Texture (SALMT) model enables clinical testing of automatic segmentation of RCT and automatic evaluation of the different tendinopathy severity levels.
השתתפות בסמינר תיתן קרדיט שמיעה = עפ"י רישום שם מלא + מספר ת.ז. בצ'ט
