Neural networks pipeline for quality management in IVF laboratory DOI Creative Commons

Sergei Sergeev,

Iuliia Diakova,

Lasha Nadirashvili

et al.

Journal of IVF-Worldwide, Journal Year: 2024, Volume and Issue: 2(4)

Published: Oct. 23, 2024

This study introduces a novel neural network-based pipeline for predicting clinical pregnancy rates in IVF treatments, integrating both and laboratory data. We developed metamodel combining deep networks Kolmogorov-Arnold networks, leveraging their complementary strengths to enhance predictive accuracy interpretability. The achieved robust performance metrics after training fitting on 11500 cases: = 0.72, AUC 0.75, F1 score 0.60, Matthews Correlation Coefficient of 0.42. According morpho-kinetical embryo evaluation, our model’s PRC 0.66 significantly improves over existing time-lapse systems prediction, demonstrating better handling imbalanced metamodel’s calibration (Brier 0.20, expected error 0.06, maximum 0.12, Hosmer-Lemeshow test p-value 0.06) indicate reliability outcomes. validated the reproducibility using an independent dataset 665 treatment cycles, showing close alignment between predicted actual (58.9% vs. 59.1%). With Bayesian method, we proposed framework historical data with real-time predictions from enabling transition retrospective prospective analysis. Our approach extends beyond conventional selection, incorporating post-analytical phase evaluation laboratory. comprehensive enables detailed analysis across different patient subpopulations time periods, facilitating identification systemic issues protocol optimization. ability track probabilities staff members allows outcome prediction assessment efficacy, providing data-driven strategy continuous improvement assisted reproductive technology.

Language: Английский

Merging synthetic and real embryo data for advanced AI predictions DOI Creative Commons
Oriana Presacan, Alexandru Dorobanțiu, Vajira Thambawita

et al.

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: March 21, 2025

Accurate embryo morphology assessment is essential in assisted reproductive technology for selecting the most viable embryo. Artificial intelligence has potential to enhance this process. However, limited availability of data presents challenges training deep learning models. To address this, we trained two generative models using datasets—one created and made publicly available, one existing public dataset—to generate synthetic images at various cell stages, including 2-cell, 4-cell, 8-cell, morula, blastocyst. These were combined with real train classification stage prediction. Our results demonstrate that incorporating alongside improved performance, model achieving 97% accuracy compared 94.5% when solely on data. This trend remained consistent tested an external Blastocyst dataset from a different clinic. Notably, even exclusively data, achieved high 92%. Furthermore, combining both yielded better than single model. Four embryologists evaluated fidelity through Turing test, during which they annotated inaccuracies offered feedback. The analysis showed diffusion outperformed adversarial network, deceiving 66.6% versus 25.3% lower Fréchet inception distance scores.

Language: Английский

Citations

0
Deep Learning-based Video Object Detection for Single-and Multi-Cell Analysis and Evaluation in Time-Lapse Imaging DOI
Taikyeong Jeong

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: May 7, 2025

Abstract Background: In this paper, we prove the efficiency of a video object detection algorithm through deep learning to have most essential time-lapse data for completion artificial intelligence vision architecture that is used prediction purpose. We alsoinvestigated data, which important part since it recorded during in vitro fertilization process. Particularly, achieve efficient by limiting special-purpose only medical healthcarebio-domains, all conditions were satisfied among single-stage videoobject architectures, and proved as theoretical proofs experiments. Method: Due characteristics bio-medical experimental purpose, applied neural networks way capture frames per second (fps)changes time-varying images. To gain advantages science mathematics biomedical domain, considered aspects entropy, confidence, occurrence probability. Accurate factors include: (i) first, accuracy number cells divided after embryo fertilization, (ii) second, acute cell size division, (iii) third, morphological uniformity embryos, (iv) fourth, possibility possible division. Results: The significant finding study accurate counting detected recognition. From an AI perspective, propose fast framework implementing evaluating two distinct models: RetinaNet, detector, Fast R-CNN, multi-stage detector. Their performance was compared against other learning-based models. Theoretical insights practical implications regarding full cycle human embryonic development derived, particularly identification abnormal temporal patterns.

Language: Английский

Citations

0
An annotated human blastocyst dataset to benchmark deep learning architectures for in vitro fertilization DOI Creative Commons
Florian Kromp,

Raphael Wagner,

Başak Balaban

et al.

Scientific Data, Journal Year: 2023, Volume and Issue: 10(1)

Published: May 11, 2023

Abstract Medical Assisted Reproduction proved its efficacy to treat the vast majority forms of infertility. One key procedures in this treatment is selection and transfer embryo with highest developmental potential. To assess potential, clinical embryologists routinely work static images (morphological assessment) or short video sequences (time-lapse annotation). Recently, Artificial Intelligence models were utilized support procedure. Even though they have proven their great potential different vitro fertilization settings, there still considerable room for improvement. advancement algorithms research field, we built a dataset consisting blastocyst additional annotations. As such, Gardner criteria annotations, depicting morphological rating scheme, collected parameters are provided. The presented intended be used train deep learning on predict Gardner’s outcomes such as live birth. A benchmark human expert’s performance annotating

Language: Английский

Citations

7
Segmentation of mature human oocytes provides interpretable and improved blastocyst outcome predictions by a machine learning model DOI Creative Commons
Jullin Fjeldstad,

Weikai Qi,

Nadia Siddique

et al.

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: May 8, 2024

Within the medical field of human assisted reproductive technology, a method for interpretable, non-invasive, and objective oocyte evaluation is lacking. To address this clinical gap, workflow utilizing machine learning techniques has been developed involving automatic multi-class segmentation two-dimensional images, morphometric analysis, prediction developmental outcomes mature denuded oocytes based on feature extraction variables. Two separate models have purpose-a model to perform multiclass segmentation, classifier classify as likely or unlikely develop into blastocyst (Day 5-7 embryo). The highly accurate at segmenting oocyte, ensuring high-quality segmented images (masks) are utilized inputs (mask model). mask displayed an area under curve (AUC) 0.63, sensitivity 0.51, specificity 0.66 test set. AUC underwent reduction 0.57 when features extracted from ooplasm were removed, suggesting holds information most pertinent competence. was further compared deep model, which also inputs. performance both combined in ensemble evaluated, showing improvement (AUC 0.67) either alone. results study indicate that direct assessments warranted, providing first insights key competence, step above current standard care-solely age proxy quality.

Language: Английский

Citations

1
Deep learning-based embryo assessment of static images can reduce the time to live birth in in vitro fertilization DOI
Yu Lu, Kevin K.W. Lam, Ernest Hung Yu Ng

et al.

Published: Oct. 29, 2024

Abstract The low success rate in vitro fertilization (IVF) may be related to our inability select embryos with good implantation potential by traditional morphology grading and remains a great challenge clinical practice. Multiple deep learning-based methods have been introduced improve embryo selection. However, existing only achieve limited prediction power generally ignore the repeated transfers from one stimulated IVF cycle. To models, we introduce Embryo2live, which assesses multifaceted qualities of static images taken under standard inverted microscope, primarily vision transformer frameworks integrate global features. We first demonstrated its superior performance predicting Gardner’s blastocyst grades up 9% improvement best method. further validated high capability supporting transfer learning using large dataset Centre. Remarkably, when applying Embryo2live for prioritization, found it improved live birth rates Top 1 patients multiple available 23.0% conventional 71.3% reducing average number 2.1 1.4 attain birth.

Language: Английский

Citations

1
Embryo Graphs: Predicting Human Embryo Viability from 3D Morphology DOI
Chloe He,

Neringa Karpavičiūtė,

Rishabh Hariharan

et al.

Lecture notes in computer science, Journal Year: 2024, Volume and Issue: unknown, P. 80 - 90

Published: Jan. 1, 2024

Language: Английский

Citations

0
Neural networks pipeline for quality management in IVF laboratory DOI Creative Commons

Sergei Sergeev,

Iuliia Diakova,

Lasha Nadirashvili

et al.

Journal of IVF-Worldwide, Journal Year: 2024, Volume and Issue: 2(4)

Published: Oct. 23, 2024

This study introduces a novel neural network-based pipeline for predicting clinical pregnancy rates in IVF treatments, integrating both and laboratory data. We developed metamodel combining deep networks Kolmogorov-Arnold networks, leveraging their complementary strengths to enhance predictive accuracy interpretability. The achieved robust performance metrics after training fitting on 11500 cases: = 0.72, AUC 0.75, F1 score 0.60, Matthews Correlation Coefficient of 0.42. According morpho-kinetical embryo evaluation, our model’s PRC 0.66 significantly improves over existing time-lapse systems prediction, demonstrating better handling imbalanced metamodel’s calibration (Brier 0.20, expected error 0.06, maximum 0.12, Hosmer-Lemeshow test p-value 0.06) indicate reliability outcomes. validated the reproducibility using an independent dataset 665 treatment cycles, showing close alignment between predicted actual (58.9% vs. 59.1%). With Bayesian method, we proposed framework historical data with real-time predictions from enabling transition retrospective prospective analysis. Our approach extends beyond conventional selection, incorporating post-analytical phase evaluation laboratory. comprehensive enables detailed analysis across different patient subpopulations time periods, facilitating identification systemic issues protocol optimization. ability track probabilities staff members allows outcome prediction assessment efficacy, providing data-driven strategy continuous improvement assisted reproductive technology.

Language: Английский

Citations

0