Deep learning segmentation of anatomical structures in three-dimensional (3D) ultrasound images of the female pelvic floor

Pelvic floor disorders (PFD) result in urinary and fecal incontinence and pelvic organ prolapse (POP), loss of support for the pelvic organs. While not commonly known, POP affects upwards of 50% of females. The common non-surgical treatment is to use intravaginal inserts, known as pessaries, to hold up the pelvic organs. The correct pessary is found using a trial-and-error method currently. Thus, there is a need to improve the pessary fitting process. To create custom pessaries, the pelvic floor must be analyzed and segmented on 3D ultrasounds in different positions because the location, size, and shape of the various structures change significantly. For this project, a deep learning-based segmentation algorithm will be developed to automatically segment the pelvic organs for improved pessary fitting, reduction in segmentation time and to avoid user variability. This project has the potential to improve the quality of life of those affected by POP.

Faculty Supervisor:

Eranga Ukwatta

Student:

Zachary Szentimrey

Partner:

Cosm Medical

Discipline:

Engineering

Sector:

Manufacturing

University:

University of Guelph