Author: Jurij Munda

Published in Applied Sciences, Volume 12, Issue 3, DOI 10.3390/app12031246

Authors: Božidar Potočnik, Martin Šavc
Faculty of Electrical Engineering and Computer Science, University of Maribor, Slovenia

Abstract: Automated detection of ovarian follicles in ultrasound images is much appreciated when its effectiveness is comparable with the experts’ annotations. Today’s best methods estimate follicles notably worse than the experts. This paper describes the development of two-stage deeply-supervised 3D Convolutional Neural Networks (CNN) based on the established U-Net. Either the entire U-Net or specific parts of the U-Net decoder were replicated in order to integrate the prior knowledge into the detection. Methods were trained end-to-end by follicle detection, while transfer learning was employed for ovary detection. The USOVA3D database of annotated ultrasound volumes, with its verification protocol, was used to verify the effectiveness. In follicle detection, the proposed methods estimate follicles up to 2.9% more accurately than the compared methods. With our two-stage CNNs trained by transfer learning, the effectiveness of ovary detection surpasses the up-to-date automated detection methods by about 7.6%. The obtained results demonstrated that our methods estimate follicles only slightly worse than the experts, while the ovaries are detected almost as accurately as by the experts. Statistical analysis of 50 repetitions of CNN model training proved that the training is stable, and that the effectiveness improvements are not only due to random initialisation. Our deeply-supervised 3D CNNs can be adapted easily to other problem domains.

Keywords: 3D Deep Neural Networks; 3D ultrasound images of ovaries; deep supervision; detection of follicles and ovaries; U-Net based architecture.

Paper in open access ( CC BY )

Published in Computer Methods and Programs in Biomedicine, Volume 196, November 2020, DOI 10.1016/j.cmpb.2020.105621

Authors: Božidar Potočnik^a, Jurij Munda^a, Milan Reljič^b, Ksenija Rakić^b, Jure Knez^b, Veljko Vlaisavljević^c, Gašper Sedej^a, Boris Cigale^d, Aleš Holobar^a, Damjan Zazula^a

^aFaculty of Electrical Engineering and Computer Science, University of Maribor, Slovenia
^bUniversity Medical Centre, Maribor, Slovenia
^cIVF ADRIA Consulting, Maribor, Slovenia
^dLogicData, Maribor, Slovenia.

Abstract: Background and objective: Automated follicle detection in ovarian ultrasound volumes remains a challenging task. An objective comparison of different follicle-detection approaches is only possible when all are tested on the same data. This paper describes the development and structure of the first publicly accessible USOVA3D database of annotated ultrasound volumes with ovarian follicles. Methods: The ovary and all follicles were annotated in each volume by two medical experts. The USOVA3D database is supplemented by a general verification protocol for unbiased assessment of detection algorithms that can be compared and ranked by scoring according to this protocol. This paper also introduces two baseline automated follicle-detection algorithms, the first based on Directional 3D Wavelet Transform (3D DWT) and the second based on Convolutional Neural Networks (CNN). Results: The USOVA3D testing data set was used to verify the variability and reliability of follicle annotations. The intra-rater overall score yielded around 83 (out of a maximum of 100), while both baseline algorithms pointed out just a slightly lower performance, with the 3D DWT-based algorithm being better, with an overall score around 78. Conclusions: On the other hand, the development of the CNN-based algorithm demonstrated that the USOVA3D database contains sufficient data for successful training without overfitting. The inter-rater reliability analysis and the obtained statistical metrics of effectiveness for both baseline algorithms confirmed that the USOVA3D database is a reliable source for developing new automated detection methods.

Keywords: 3D Ultrasound images of ovaries, Detection of ovarian follicles, Public database, Unbiased verification of detection algorithms, Web services.

Highlights:

• Publishing of the USOVA3D public database of annotated 3D ovarian ultrasound images.
• Ovaries and follicles annotated by two gynaecologists.
• Design of a verification protocol for unbiased assessment of detection algorithms.
• Introduction of two advanced algorithms for follicle and ovary detection.
• Inter-rater variability and baseline performance assessed on this database.

Free access for the article (al least) until August 18, 2020!