Chun-I Lee1,2,3, Yan-Ru Su4, Chien-Hong Chen3, T Arthur Chang5, Esther En-Shu Kuo4, Wei-Lin Zheng4, Wen-Ting Hsieh4, Chun-Chia Huang3, Maw-Sheng Lee1,2,3, Mark Liu6. 1. Institute of Medicine, Chung Shan Medical University, Taichung, Taiwan. 2. Department of Obstetrics and Gynecology, Chung Shan Medical University, Taichung, Taiwan. 3. Division of Infertility, Lee Women's Hospital, Taichung, Taiwan. 4. Binflux Inc., Taipei, Taiwan. 5. Department of Obstetrics and Gynecology, University of Texas Health Science Center, San Antonio, TX, USA. 6. Binflux Inc., Taipei, Taiwan. markliu@binflux.com.
Abstract
PURPOSE: Our retrospective study is to investigate an end-to-end deep learning model in identifying ploidy status through raw time-lapse video. METHODS: By randomly dividing the dataset of time-lapse videos with known outcome of preimplantation genetic testing for aneuploidy (PGT-A), a deep learning model on raw videos was trained by the 80% dataset, and used to test the remaining 20%, by feeding time-lapse videos as input and the PGT-A prediction as output. The performance was measured by an average area under the curve (AUC) of the receiver operating characteristic curve. RESULT(S): With 690 sets of time-lapse video image, combined with PGT-A results, our deep learning model has achieved an AUC of 0.74 from the test dataset (138 videos), in discriminating between aneuploid embryos (group 1) and others (group 2, including euploid and mosaic embryos). CONCLUSION: Our model demonstrated a proof of concept and potential in recognizing the ploidy status of tested embryos. A larger scale and further optimization on the exclusion criteria would be included in our future investigation, as well as prospective approach.
PURPOSE: Our retrospective study is to investigate an end-to-end deep learning model in identifying ploidy status through raw time-lapse video. METHODS: By randomly dividing the dataset of time-lapse videos with known outcome of preimplantation genetic testing for aneuploidy (PGT-A), a deep learning model on raw videos was trained by the 80% dataset, and used to test the remaining 20%, by feeding time-lapse videos as input and the PGT-A prediction as output. The performance was measured by an average area under the curve (AUC) of the receiver operating characteristic curve. RESULT(S): With 690 sets of time-lapse video image, combined with PGT-A results, our deep learning model has achieved an AUC of 0.74 from the test dataset (138 videos), in discriminating between aneuploid embryos (group 1) and others (group 2, including euploid and mosaic embryos). CONCLUSION: Our model demonstrated a proof of concept and potential in recognizing the ploidy status of tested embryos. A larger scale and further optimization on the exclusion criteria would be included in our future investigation, as well as prospective approach.
Authors: Ermanno Greco; Katarzyna Litwicka; Maria Giulia Minasi; Elisabetta Cursio; Pier Francesco Greco; Paolo Barillari Journal: Int J Mol Sci Date: 2020-06-19 Impact factor: 5.923