Convolutional neural network-based automatic detection of follicle cells in ovarian tissue using optical coherence tomography. (3rd November 2020)

• Record Type: Journal Article
• Authors: Saito, Kasumi; Motani, Yuki; Takae, Seido; Suzuki, Nao; Tsukada, Kosuke
• Is Part Of: Biomedical physics & engineering express. Volume 6:Number 6(2020)
• Publication Date: 2020-11-03
• Subjects: optical coherence tomography -- image analysis -- convolutional neural network -- ovarian tissue; Medical physics -- Periodicals; Biophysics -- Periodicals; Biomedical engineering -- Periodicals; Medical sciences -- Periodicals; 610.153
• DOI: 10.1088/2057-1976/abc3d4
• Languages: English
• ISSNs: 2057-1976

Abstract

Abstract: To preserve the fertility of young female cancer patients, ovarian tissue cryopreservation and transplantation have been investigated as next-generation reproductive medical technologies. Non-invasive visualization of follicles in ovarian tissue and cryopreservation of higher density tissue is essential for effective transplantation. We proposed the use of optical coherence tomography (OCT) that can noninvasively visualize the internal structure of the ovarian tissue. However, a method for quantifying cell density has not yet been established because of the lack of available techniques to visualize follicles noninvasively. We proposed the use of a convolutional neural network (CNN) to extract small features from medical images as an image analysis method to automatically detect follicles from the obtained OCT images. First, we collected a total of 13 ovarian tissues from four-day-old mice and acquired OCT images using a full-field-type OCT. Then, the acquired images were analyzed using three detection methods: filter processing, filter processing combined with the CNN, and only CNN. Finally, to verify the detection accuracy of each method, the detection rate and precision were calculated by taking the doctor's detection as the correct result. The results showed that the detection method only using CNN achieved a detection rate of 0.81 and precision of 0.67; this indicated that follicles could be effectively detected using our proposed method. Furthermore, it is quantitatively evident that the density of follicles from the surface layer to the deep region differs depending on the tissue. In the future, these results could be used to detect follicles in tissues of different maturation stages and quantify follicles three-dimensionally, further accelerating next-generation reproductive medicine.