Pd37-03 Deepsperm: A Machine Learning Based Model That Facilitates Sperm Identification Through Touch Print Smear Cytology of Testicular Specimen From Patients With Azoospermia

You have accessJournal of UrologyInfertility: Epidemiology & Evaluation II (PD37)1 May 2024PD37-03 DEEPSPERM: A MACHINE LEARNING BASED MODEL THAT FACILITATES SPERM IDENTIFICATION THROUGH TOUCH PRINT SMEAR CYTOLOGY OF TESTICULAR SPECIMEN FROM PATIENTS WITH AZOOSPERMIA Chen-Hao Hsu, Cheng-Han Tsai, Mong-Chi Ko, I-Shen Huang, Wei-Jen Chen, Yu-Ching Peng, Chun-Fu Yeh, Tyng-Luh Liu, Ethan Tu, and William J. Huang Chen-Hao HsuChen-Hao Hsu , Cheng-Han TsaiCheng-Han Tsai , Mong-Chi KoMong-Chi Ko , I-Shen HuangI-Shen Huang , Wei-Jen ChenWei-Jen Chen , Yu-Ching PengYu-Ching Peng , Chun-Fu YehChun-Fu Yeh , Tyng-Luh LiuTyng-Luh Liu , Ethan TuEthan Tu , and William J. HuangWilliam J. Huang View All Author Informationhttps://doi.org/10.1097/01.JU.0001009464.98066.03.03AboutPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissionsReprints ShareFacebookLinked InTwitterEmail Abstract INTRODUCTION AND OBJECTIVE: Testicular touch print smear (TPS) cytology provides information about spermatogenesis during microdissection testicular sperm extraction (mTESE) for sperm retrieval among azoospermic patients. We propose a machine learning (ML) based model, DeepsperM, to facilitate TPS reading. METHODS: 147 microscopic TPS images obtained from the testicular specimen of patients with azoospermia were retrospectively collected between 2019 and 2021. The cells on each image including Sertoli cell, primary spermatocytes, round spermatids, elongated spermatids, immature sperm, and mature sperm were annotated by three clinicians with bounding boxes. Among them, 118 images were assigned as the training set and 29 images as the validation set. RetinaNet, a one-stage detection framework, was adopted for cell detection and classification. The performance was evaluated at the cell level and advanced grouping of cell types with average precision (AP) and recall. The precision-recall (PR) curve was displayed and compared with three clinicians among a testing set that contains another 29 images. RESULTS: The training set consisted of 4772 annotated cells, including 1782 Sertoli cells, 314 primary spermatocytes, 443 round spermatids, 279 elongated spermatids, 504 immature sperm, and 1450 mature sperm. The overall AP and recall on the validation set with 6 categories of cells were 80.47% and 96.69%, and they were 79.48% and 93.63% on the testing set, respectively. The performance of the overall AP and recall increased to 85.29% and 93.80% once the post-meiotic cells including round spermatids, elongated spermatids, immature sperm and mature sperm were merged into one category, indicating advanced grouping of cell types that share similar characteristics may elevate the performance. The PR curve exhibited comparable performance of DeepsperM with the participating clinicians. CONCLUSIONS: This study proposed an innovative approach leveraging ML methods to facilitate the diagnosis of spermatogenesis at mTESE for patients with azoospermia. With the assistance of DeepSperM, clinicians could overcome the steep learning curve of TPS interpretation and provide more rapid detection of post-meiotic cells for further assisted reproductive technologies. Source of Funding: Taipei Veterans General Hospital (VGH109-C-132 and VGH112-C-159); National Science and Technology Council of Taiwan (109-2314-B-075-073-MY2) © 2024 by American Urological Association Education and Research, Inc.FiguresReferencesRelatedDetails Volume 211Issue 5SMay 2024Page: e800 Advertisement Copyright & Permissions© 2024 by American Urological Association Education and Research, Inc.Metrics Author Information Chen-Hao Hsu More articles by this author Cheng-Han Tsai More articles by this author Mong-Chi Ko More articles by this author I-Shen Huang More articles by this author Wei-Jen Chen More articles by this author Yu-Ching Peng More articles by this author Chun-Fu Yeh More articles by this author Tyng-Luh Liu More articles by this author Ethan Tu More articles by this author William J. Huang More articles by this author Expand All Advertisement PDF downloadLoading ...