Automated ovarian volume quantification in transvaginal ultrasound
- Resource Type
- Conference
- Authors
- Narra, Ravi Teja; Singhal, Nitin; Narayan, Nikhil S.; Ramaraju, G A
- Source
- 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018) Biomedical Imaging (ISBI 2018), 2018 IEEE 15th International Symposium on. :1513-1516 Apr, 2018
- Subject
- Bioengineering
Image segmentation
Two dimensional displays
Volume measurement
Ultrasonic imaging
Three-dimensional displays
Training
Image edge detection
Ovary
Ultrasound
Segmentation
U-V parametrization
Deep-Learning
- Language
- ISSN
- 1945-8452
Ovarian quantification by volume measurement is performed in routine clinical practice for diagnosis and management of gynecological conditions such as infertility. This paper describes an automated algorithm for ovarian volume measurement using three-dimensional transvaginal ultrasound (TVUS) images. The algorithm integrates deep learned energy map as a soft shape prior within the variational segmentation framework for 2D radial slice segmentation. The output of the segmentation framework is used for mesh generation using U-V spherical parametrization to estimate the surface of the ovarian volume. The segmentation framework provides approximately 86% average Dice overlap with the ground truth annotations. The mean absolute volume difference is found to be approximately 5ml between manual and automated measurements on 55 TVUS images.