| Resumo: | Infertility affects about 186 million people worldwide and 9-10% of couples in Portugal, causing financial, social and medical problems. Evaluation of embryo quality based morphological features is the standard in vitro fertilization (IVF) clinics around the world. This process is subjective and time-consuming, and results in discrepant classifications among embryologists and clinics, leading to fail in predict accurately embryo implantation and live birth potential. Although assisted reproductive technologies (ART) such as IVF coupled with time lapse elimination of periodic transfer to microscopy assessment and stable embryo culture conditions for embryos development, has alleviated the infertility problem, there are significant limitations even considering morphokinetic analysis. Likewise, many patients require multiple IVF cycles to achieve pregnancy, making the selection of single embryo for transfer a critical challenge. Here, we demonstrate the reliability of machine learning, especially deep learning based on TensorFlow open source and Keras libraries for embryo raw TLI images features extraction and classification in clinical practice. Equally, we present a follow up pipeline for clinicians and researchers, with no expertise in machine learning, to easily, rapid and accurately utilize deep learning as a clinical decision support tool in embryos viability studies, as well in other medical field where the analysis of images is preeminent
|
|---|