Predicting Live Birth After Fresh Embryo Transfer in IVF/ICSI: Differential ICM/Trophectoderm Morphology Encoding and a Hybrid Attention--Logistic Regression Ensemble

Abdul Raheem Shahzad; Jinxin Chen; Ming Zhang; Juan Deng

cs.LGOct 15, 2026

Predicting Live Birth After Fresh Embryo Transfer in IVF/ICSI: Differential ICM/Trophectoderm Morphology Encoding and a Hybrid Attention--Logistic Regression Ensemble

Authors: Abdul Raheem Shahzad, Jinxin Chen, Ming Zhang, Juan Deng

Organizations: School of Computer Science, Wuhan University, Wuhan, China · Department of Obstetrics and Gynecology, Zhongnan Hospital of Wuhan University, Wuhan, China · Reproductive Medical Center, Zhongnan Hospital of Wuhan University, Wuhan, Hubei, P.R. China, and also with the Clinical Medicine Research Center of Prenatal Diagnosis and Birth Health in Hubei Province, Wuhan, Hubei, P.R. China

Abstract

Predicting live birth before fresh embryo transfer could support shared decision-making in IVF/ICSI, but routine clinical data are small, heterogeneous, and contain pathway-dependent variables prone to outcome leakage. We aimed to construct a curated single-centre IVF/ICSI dataset and develop a prediction model suited to small tabular clinical data. Materials and Methods: We assembled 616 fresh embryo-transfer cycles (251 live birth, 365 non-live birth) with clinical, laboratory, and embryological predictors, and embryo morphology encoded with a novel differential weighting of the inner cell mass and trophectoderm. After KNN imputation, robust scaling, and selection of 50 features, we developed a Hybrid Attention Network (HAN), a learnable feature-attention gate followed by a compact multilayer perceptron, and a HAN + logistic-regression (LR) ensemble with a validation-tuned weight. Models were benchmarked against classical machine learning, TabPFN, and text-serialized language models on a held-out test set; AUC-ROC was the primary metric. Results: The HAN + LR ensemble achieved the highest test AUC-ROC (0.710; accuracy 0.710, F1 0.630), ahead of the HAN backbone alone (0.700), logistic regression (0.667), the TabPFN + LR hybrid (0.666), and TabPFN (0.650). Text-serialized language models performed near chance (AUC $\le$ 0.61). The differentially weighted trophectoderm feature ranked highest by consensus importance across models. Discussion: A compact attention network combined with a linear model outperformed both components and strong foundation-model baselines on a small clinical cohort. Conclusion: The curated dataset and HAN + LR ensemble offer a transparent, reproducible basis for IVF/ICSI live-birth prediction; external validation is the next step.

Figures & tables

Variable	Live birth (n = 251)	Non-live birth (n = 365)	$p$	SMD	Selected
Female factors
Female age (years)	31.8 $\pm$ 3.7	33.1 $\pm$ 4.0	$<$ 0.001	0.34	$\checkmark$
Female smoking history, n (%)			0.722	–	–
Have	4 (1.6%)	4 (1.1%)
None	247 (98.4%)	361 (98.9%)
Female height (cm)	160.8 $\pm$ 4.9	161.2 $\pm$ 4.5	0.269	0.09	$\checkmark$

TABLE I: Baseline Characteristics of the Study Cohort by Outcome Group, With the Modelling Status of Every Collected Variable

Variable	Live birth (n = 251)	Non-live birth (n = 365)	$p$	SMD	Selected
Stimulation and laboratory
Ovarian stimulation protocol, n (%)			0.795	–	$\checkmark$
Antagonist regimen	199 (79.3%)	296 (81.1%)
Extra Long Plan	3 (1.2%)	7 (1.9%)
Long protocol for early follicular phase	21 (8.4%)	29 (7.9%)
Luteal phase extension protocol	26 (10.4%)	29 (7.9%)

TABLE I: (Continued)

Fig. 1: Diagram of the embryo morphology encoding scheme.

Component	Setting
Feature Attention Module	Learnable per-feature gate, sigmoid, element-wise multiplicative
MLP layers	$50\rightarrow 24$ (LayerNorm, ReLU, dropout 0.25); $24\rightarrow 12$ (ReLU, dropout 0.15); $12\rightarrow 1$ (sigmoid)
HAN ensemble	3 seeds (42, 123, 456), probabilities averaged
Parameters	$\approx$ 1,635 per HAN instance; 4,957 total
Ensemble rule	$p=w\,p_{\mathrm{HAN}}+(1-w)\,p_{\mathrm{LR}}$ ; $w=0.70$ by validation grid search
Optimizer / loss	AdamW (lr 0.002, weight decay 0.001); binary cross-entropy

TABLE II: HAN + LR Architecture and Training Configuration

Model	AUC	Acc	Prec	Rec	F1
HAN + LR (proposed)	0.710	0.710	0.657	0.605	0.630
HAN backbone (alone)	0.700	0.667	0.590	0.605	0.597
Logistic regression	0.667	0.645	0.558	0.632	0.593
TabPFN + LR	0.666	0.656	0.615	0.421	0.500
TabPFN	0.650	0.613	0.545	0.316	0.400
Support-vector machine	0.632	0.624	0.588	0.263	0.364

TABLE III: Test-Set Performance of the Eleven Evaluated Models, Ranked by AUC-ROC

Fig. 2: Receiver-operating-characteristic (AUC-ROC) curves for the evaluated models on the held-out test set.

Feature	Domain	Consensus (%)	RF	GB	XGB	LR	HAN
Blastocyst 1 TE (weighted)	Oocyte retrieval and embryo development	4.53	0.008	0.019	0.014	1.208	0.537
Ovarian stimulation protocol (Extra Long Plan)	Stimulation and laboratory	3.35	0.001	0.000	0.000	0.759	0.560
Method of fertilization (IVF+RICSI)	Stimulation and laboratory	2.93	0.001	0.000	0.000	0.599	0.555
Cleavage embryo 1 grade	Oocyte retrieval and embryo development	2.66	0.003	0.000	0.001	0.523	0.520
Number of follicles punctured	Oocyte retrieval and embryo development	2.61	0.024	0.007	0.009	0.464	0.522
Number of oocytes retrieved	Oocyte retrieval and embryo development	2.57	0.020	0.019	0.015	0.427	0.532

TABLE IV: Top 15 Predictors by Consensus Importance, With the Native Importance Reported by Each Model

Fig. 3: Top 15 predictors by consensus importance, averaged across the five models that expose an importance measure and coloured by the clinical domains defined in Table I .

Fig. 4: Relative importance of each of the top 15 predictors within each model, min-max scaled per model so that differently scaled measures can be compared. Scaling amplifies the narrow spread of the HAN attention weights and should be read as ordering, not magnitude.

Fig. 5: Share of total consensus importance contributed by each of the four clinical domains defined in Table I , with the number of selected features in each domain.

Clinical Outcome Prediction

Clinical outcome prediction uses patient records, imaging, signals, and biomarkers to estimate risks, disease trajectories, treatment response, and survival.

See all 265 papers

InVitroVision: a Multi-Modal AI Model for Automated Description of Embryo Development using Natural Language
Apr 22, 2026Nicklas Neu, Thomas Ebner, Jasmin Primus +4VLM Adaptation Medical VLMs
Software Competence Center Hagenberg GmbH,Softwarepark 32a,Hagenberg, 4232, Upper Austria, Austria · Kinderwunsch Zentrum Kepler Universitätsklinikum,Krankenhausstraße 26, Linz, 4021 Upper Austria, Austria · Wunschkind Klinik Dr. Brunbauer, Ebendorferstraße 6/4 Vienna, 1010, Vienna, Austria
Predicting Blastocyst Formation in IVF: Integrating DINOv2 and Attention-Based LSTM on Time-Lapse Embryo Images
Apr 14, 2026Zahra Asghari Varzaneh, Niclas Wölner-Hanssen, Reza Khoshkangini +2Microscopy Image Analysis Recurrent Neural Networks
Sustainable Digitalisation Research Center, Department of Computer Science and Media Technology, Malmö University, Malmö, Sweden · School of Information Technology, Halmstad University, Halmstad, Sweden · Kepler Universitätsklinikum, Linz, Austria +1
Expert-Annotated Embryo Image Dataset with Natural Language Descriptions for Evidence-Based Patient Communication in IVF
Apr 16, 2026Nicklas Neu, Thomas Ebner, Jasmin Primus +4Medical VLMs Explainable Medical Image Analysis
1Software Competence Center Hagenberg, Hagenberg, 4232, Austria · 2Kepler Universitätsklinikum, Kinderwunsch Zentrum, Linz, 4020, Austria · 3Wunschkind Klinik Dr. Brunbauer, Wien, 1010, Austria

Apr 22, 2026J/K move · Enter open · S save

InVitroVision: a Multi-Modal AI Model for Automated Description of Embryo Development using Natural Language

Nicklas Neu, Thomas Ebner, Jasmin Primus, Raphael Zefferer+3

Software Competence Center Hagenberg GmbH,Softwarepark 32a,Hagenberg, 4232, Upper Austria, Austria · Kinderwunsch Zentrum Kepler Universitätsklinikum,Krankenhausstraße 26, Linz, 4021 Upper Austria, Austria · Wunschkind Klinik Dr. Brunbauer, Ebendorferstraße 6/4 Vienna, 1010, Vienna, Austria

VLM Adaptation Medical VLMs Microscopy Image Analysis

Open paper AskPDF arXiv

The application of artificial intelligence (AI) in IVF has shown promise in improving consistency and standardization of decisions, but often relies on annotated data and does not make use of the multimodal nature of IVF data. We investigated whether foundational vision-language models can be fine-tuned to predict natural language descriptions of embryo morphology and development. Using a publicly available embryo time-lapse dataset, we fine-tuned PaliGemma-2, a multi-modal vision-language model, with only 1,000 images and corresponding captions, describing embryo morphology, embryonic cell cycle and developmental stage. Our results show that the fine-tuned model, InVitroVision, outperformed a commercial model, ChatGPT 5.2, and base models in overall metrics, with performance improving with larger training datasets. This study demonstrates the potential of foundational vision-language models to generalize to IVF tasks with limited data, enabling the prediction of natural language descriptions of embryo morphology and development. This approach may facilitate the use of large language models to retrieve information and scientific evidence from relevant publications and guidelines, and has implications for few-shot adaptation to multiple downstream tasks in IVF.

Abstract

Clinical Outcome Prediction

Explore similar work

InVitroVision: a Multi-Modal AI Model for Automated Description of Embryo Development using Natural Language

Predicting Blastocyst Formation in IVF: Integrating DINOv2 and Attention-Based LSTM on Time-Lapse Embryo Images

Expert-Annotated Embryo Image Dataset with Natural Language Descriptions for Evidence-Based Patient Communication in IVF