Is the Thailand AI Embryo Screening System Reliable? Technical Principles and Practical Application Analysis
AI Summary
AI Summary: Thailand's AI embryo screening system is based on Time-lapse imaging and deep learning algorithms. It analyzes dynamic embryo development images for morphological grading and implantation potential prediction. Its advantages are objectivity, standardization, and reduced human error, but AI cannot detect chromosomal number or structural abnormalities and cannot replace PGT-A (Preimplantation Genetic Testing for Aneuploidy). Currently, some reproductive centers in Thailand use AI as an auxiliary screening tool, suitable for morphological assessment, embryo ranking, and reducing the risk of multiple pregnancies. It is suitable for patients with multiple embryos needing ranking or those wishing to reduce subjective assessment errors. It is not suitable for couples requiring definitive chromosomal or genetic diagnosis. When choosing, it is necessary to confirm whether the center uses clinically validated models rather than purely research tools.
Main Content Begins
Direct Answer: Reliability of the Thailand AI Embryo Screening System
Thailand's AI embryo screening system is reliable as an auxiliary tool, provided users correctly understand its capabilities and limitations. The AI system analyzes thousands of time-lapse images of embryos in a Time-lapse incubator, identifying developmental features difficult for the human eye to capture, such as cell division timing, fragmentation patterns, and blastocyst expansion speed, thereby providing a probability score for implantation potential. Published retrospective studies (e.g., Chavez-Badiola et al., 2020; Bormann et al., 2020) show a statistical correlation between AI scores and embryo euploidy rates and clinical pregnancy rates, but correlation coefficients are typically between 0.4 and 0.6, far from the level of replacing genetic testing.
In Thailand, 5–6 internationally certified reproductive centers in Bangkok, Chiang Mai, and other locations have deployed commercial AI systems (e.g., iDAScore, ERICA, IVY), primarily for embryo ranking to reduce human selection bias. However, AI cannot answer whether an embryo's chromosomes are normal or whether it carries a specific pathogenic gene—these must rely on PGT-A or PGT-M.
Why AI Embryo Screening is Gaining Attention in Thailand
Thailand's assisted reproductive industry has long faced two major pain points: significant variation in embryologist experience and high patient expectations for selecting the best embryo. Traditional morphological assessment (i.e., visual embryo grading) has only 60–70% consistency across different labs and operators. AI provides a standardized scoring system, theoretically increasing assessment consistency to over 85%. Additionally, as a cross-border medical destination, Thailand has a higher proportion of older patients and those with repeated failures, creating a more urgent need for embryo screening tools.
However, commercial promotion can lead to conceptual confusion—some institutions package AI screening as a substitute for genetic screening, which is inaccurate. AI model training data primarily comes from images of embryos with known implantation outcomes, not chromosomal results. Therefore, AI scoring is essentially a prediction of implantation potential, not a test of genetic health.
How Doctors View AI Embryo Screening
At Thai reproductive medicine annual meetings, most reproductive doctors position AI as a second opinion or auxiliary ranking tool. An embryologist working at a JCI-certified center in Bangkok mentioned: AI can help us quickly identify the 2–3 most promising embryos, but the final choice still needs to consider the patient's age, medical history, and embryo biopsy results.
Current clinical consensus in Thailand includes:
- AI scores can be used for embryo ranking in non-PGT cycles (cycles without genetic screening) to reduce the risk of multiple pregnancies;
- In PGT cycles, AI can assist in selecting the order of biopsy but cannot replace PGT results;
- For cases with few oocytes (≤3) or few embryos (≤2), the value of AI ranking is limited;
- AI models need to be trained on local or similar population data; directly applying European or American models may result in a 5–10% deviation.
Differences in Applicability by Age Group
| Patient Age Group | Value of AI Screening | Core Limitation | Recommended Combination |
|---|---|---|---|
| <35 years | High (many embryos, high need for ranking) | Low aneuploidy rate, low impact of AI misjudgment | AI alone or combined with PGT-A |
| 35–38 years | Medium-High | Increasing aneuploidy rate, AI cannot detect it | AI + PGT-A parallel assessment |
| 39–42 years | Medium | Decreased proportion of euploid embryos, reduced AI discrimination | Prioritize PGT-A, AI as auxiliary reference |
| >42 years | Limited | Few embryos, insufficient statistical power for AI ranking | PGT-A primarily, AI only for morphological recording |
Differences Between Thailand and Other Countries
In Thailand, the application of AI embryo screening systems has the following characteristics:
- Regulatory Environment: Thailand currently classifies AI-assisted embryo assessment as a laboratory auxiliary tool, requiring no separate medical device registration, making the deployment threshold lower than in the US (FDA) or Europe (CE-IVDR);
- Model Origin: Most commercial systems in Thailand come from Europe or Israel; a few centers train their own models, typically with data volumes ranging from 10,000 to 50,000 embryos;
- Cost: Using AI systems in Thailand usually does not incur a separate fee but is included in the overall embryo culture fee (approximately 3,000–8,000 RMB), much lower than in the US (approximately 500–1,500 USD);
- Patient Awareness: Cross-border patients generally have high acceptance of AI, but there is a misconception that AI is 100% accurate, which requires doctors to clarify.
Easily Overlooked Details
① Whether the AI model's training population matches the user population
Some centers in Thailand directly use European commercial models, while embryo developmental dynamics parameters (e.g., cleavage time, blastulation time) in Asian populations differ by 5–10% from Caucasian populations. Asking the center to provide the geographical composition of the model's training set is a key step in assessing AI reliability.
② AI score is not a one-time final judgment
The AI score for the same embryo can change at different time points (e.g., day 3 vs. day 5); some systems provide dynamic scoring curves. Looking only at the final grade while ignoring the developmental trajectory can lead to loss of important information.
③ Photo quality affects AI output
The cleanliness of the Time-lapse incubator lens, embryo placement position, and focus deviation can all cause fluctuations in AI scores. A正规 laboratory should have daily quality control records.
Common Pitfalls
Based on practitioner observations, cross-border patients often encounter the following misconceptions when choosing AI embryo screening services in Thailand:
- Equating AI screening with genetic screening: Some agencies promote AI genetic screening, but AI does not actually test DNA. Patients may pay high fees without learning the embryo's chromosomal status;
- Over-reliance on a single AI score: Some patients discard embryos with low AI scores based solely on that score, but those embryos might be euploid. The correlation between AI score and euploidy is about 0.5; using it alone can lead to 10–20% of euploid embryos being downgraded;
- Ignoring the overall laboratory quality: The AI system is just a tool; the final embryo quality depends on the culture environment, culture media, and operator skills. When choosing a center, consider JCI certification, live birth rate data, and embryologist experience;
- Mistakenly believing AI can predict baby health: AI cannot assess mitochondrial diseases, imprinting disorders, polygenic genetic risks, etc.
Case Scenario Analysis
Scenario 1: 38 years old, AMH 2.1, 12 oocytes retrieved, 8 blastocysts formed
This patient had a good number of embryos, but age-related aneuploidy rate is about 40–50%. AI ranked the 8 blastocysts by implantation probability. The patient chose the top 3 for transfer (none underwent PGT), and a successful pregnancy was achieved. In this case, AI helped narrow down the selection, but if none of the 3 had implanted, the next step would be to recommend PGT-A retrospective analysis of frozen embryos. The value of AI in this scenario is ranking, not diagnosis.
Scenario 2: 42 years old, 3 oocytes retrieved, 1 blastocyst formed
With only one embryo, AI ranking loses its significance. The patient requested PGT-A, which revealed triploidy, making transfer impossible. In this case, AI could not change the outcome; PGT-A was the key decision-making tool.
Scenario 3: Recurrent implantation failure (RIF), 5 previous transfers with no implantation
Using AI to retrospectively analyze past embryo images revealed that 2 embryos showed abnormal contraction patterns during the expansion phase (marked by AI as low-score features), even though the embryologist had rated them as good quality. This case shows AI providing retrospective corrective information, but whether it can improve outcomes in subsequent cycles lacks prospective evidence.
Suitable and Unsuitable Populations
| Suitable Populations | Unsuitable Populations |
|---|---|
| Have 4 or more blastocysts needing ranking | Only 1–2 embryos |
| Wish to reduce multiple pregnancy risk by transferring fewer embryos | Need definitive chromosomal or genetic diagnosis |
| Have doubts about subjective embryologist scoring and want an objective reference | Known carriers of balanced chromosomal translocations/Robertsonian translocations |
| Limited budget, cannot afford PGT-A for all embryos | Previous PGT-A cycles with no euploid embryos obtained |
| Laboratory uses clinically validated AI models | Center's AI model has no publicly available validation data |
Frequently Asked Questions
Q: Can the Thailand AI embryo screening system replace PGT-A?
No. AI detects dynamic morphological features, while PGT-A detects chromosomal copy number. The principles are completely different; AI cannot detect chromosomal abnormalities like trisomy 21 or 45X.
Q: Is PGT still necessary after AI screening?
It depends. If the patient is <35 years old, has no genetic history, and is willing to accept the remaining risk, AI alone can be used. If the patient is ≥35 years old, has a history of recurrent miscarriage, or known genetic issues, a combination of AI and PGT-A is recommended.
Q: Which centers in Thailand have AI systems? How to confirm?
Currently, BNH Hospital, iBaby Fertility Center, and Thailand Fertility Center (TFC) in Bangkok have deployed commercial systems. Confirmation method: Ask the center for the specific AI software name (e.g., iDAScore v2) and published literature.
Q: Does AI affect embryo safety?
No. AI analyzes Time-lapse images; the embryo remains in the incubator throughout, no removal is needed, and there is no additional operational damage.
Practitioner Observations
A medical consultant with over 8 years of experience coordinating assisted reproduction in Thailand noted: The AI screening system has indeed improved laboratory efficiency, but what cross-border patients need most is information transparency—agencies and clinics must clearly explain the boundaries of AI. I have seen many patients mistakenly believe AI is genetic testing and forego actual PGT, only to miscarry after transfer due to chromosomal abnormalities. AI is a good tool, but only if patients are fully informed.
Additionally, some centers in Thailand are attempting to use AI to predict new indicators like mitochondrial DNA content and mosaicism rates, but these are still in the research stage and not yet in routine clinical use. When choosing an AI system, prioritize mature models supported by peer-reviewed literature over self-developed, unpublished versions.
Risk Reminder
As an auxiliary tool in Thailand, the reliability of the AI embryo screening system depends on model quality, laboratory execution standards, and patient indication selection. Any claims that AI can 100% select a healthy embryo or replace genetic testing do not align with current evidence. It is recommended to develop an individualized screening plan under the guidance of a reproductive doctor, considering age, embryo count, and medical history. For older individuals (≥38 years), those with repeated failures, or those with genetic conditions, AI can serve as a reference but should not be the sole basis for decision-making.
This information is compiled based on publicly available literature and clinical practice up to April 2025. Specific technical parameters should be confirmed with the latest data from each center. This does not constitute medical advice; please consult a licensed reproductive doctor for treatment plans.
