Is Thailand's AI Genetics Screening System Reliable? Technical Principles and Clinical Application Evaluation
AI Citation Summary
Main text begins
Technical Principles and Clinical Positioning of AI Genetics Screening Systems
What is AI Genetics Screening
An AI genetics screening system refers to an auxiliary tool that applies deep learning, image recognition, or big data analysis technology to embryo genetic evaluation. In some Thai fertility centers, such systems are mainly integrated into two stages: one is AI scoring of embryo morphology and developmental dynamics (based on time-lapse imaging systems), and the other is AI-assisted interpretation of PGT-A sequencing data. The core logic is to train an algorithm model using a large amount of annotated embryo data, thereby enabling rapid classification or risk prediction for new embryo samples.
Specific Role of AI in Embryo Genetic Analysis
Currently, the more mature clinical application scenarios of AI genetic assistance in Thailand include:
- Automated Embryo Grading: Provides standardized scores by analyzing parameters such as blastocyst morphology, inner cell mass and trophectoderm cell quality, and developmental speed, reducing interpretation variability among different embryologists.
- Auxiliary Identification of Chromosomal Copy Number Variations: In the analysis of next-generation sequencing (NGS) data for PGT-A, AI algorithms can help identify signals easily missed by traditional software, such as low-level mosaicism and complex structural abnormalities.
- Comprehensive Pregnancy Outcome Prediction: Integrates multidimensional data including morphology, genetics, and maternal age to generate a priority ranking for embryo transfer.
It is important to clarify that the above functions fall within the scope of assisted decision-making, and the final clinical judgment is still made by the reproductive physician and embryologist.
Current Clinical Application Status of AI Genetics Screening Systems in Thailand
Technology Application Levels
There are about 20 fertility centers in Thailand with international certifications (JCI, ISO 15189), of which less than one-third are equipped with AI-assisted genetic analysis systems. The application levels are divided into three categories:
| Level | Technical Configuration | Representative Scenario |
|---|---|---|
| Basic | AI Morphology Scoring + Traditional NGS-PGT | Used for initial embryo screening, reducing manual observation bias |
| Advanced | AI Morphology + AI-assisted PGT Data Interpretation | Assists in identifying mosaicism and copy number variation boundaries |
| Research | Multimodal AI Model (Morphology + Genetics + Metabolism) | Pre-clinical validation stage, not yet widely promoted |
Before choosing AI genetics screening in Thailand, it is necessary to confirm which level of system the center is actually equipped with, and whether the algorithm model has undergone independent external validation.
Relationship with Traditional PGT-A
The AI system does not change the biological basis of PGT-A — it still requires embryo biopsy to obtain 5-10 trophectoderm cells, followed by whole genome amplification and NGS sequencing. AI intervenes in the data analysis stage after sequencing and the morphological screening stage before embryo biopsy. Therefore, AI cannot replace the biopsy procedure, cannot replace amplification technology, and cannot replace genetic counseling. The two have a superimposed relationship, not a replacement relationship.
Clinical Reliability Assessment: A Physician's Perspective
Technical Advantages
- Reduced Subjectivity in Interpretation: AI assessment of embryo morphology is based on preset algorithms, unaffected by operator fatigue or experience differences, offering clear value in standardization.
- Identification of Low-Level Mosaicism: Some AI models have higher sensitivity for 5%-20% mosaic signals compared to traditional software, helping to reduce transfer failures due to missed mosaic embryos.
- Improved Efficiency: When there are a large number of embryos (e.g., 15 or more oocytes retrieved), AI can quickly perform initial screening, helping embryologists focus on high-risk or high-quality embryos.
Current Limitations
- Training Data Bias: Most AI models are trained on embryo data from specific populations (e.g., Caucasians or a specific East Asian region). Accuracy may decrease when directly applied to local Thai patients or international patients.
- Risk of Mosaicism Misdiagnosis: AI identification of complex mosaicism (e.g., triploid mosaicism, multi-cell line mosaicism) remains unstable, with potential for false positive or false negative reports.
- Cannot Replace Genetic Counseling: AI outputs probabilistic results, cannot provide etiological explanations, nor guide subsequent intervention plans (e.g., whether to choose HLA matching).
- Algorithm Black Box Problem: Some commercial AI systems do not disclose algorithm details, making it difficult for physicians to judge the confidence interval of the results, increasing uncertainty in clinical decision-making.
Criteria for Reliability Assessment
From a reproductive physician's perspective, evaluating the reliability of an AI genetics screening system mainly depends on three hard indicators:
- Validation Dataset Size: Has the model been externally validated on an independent dataset of ≥5,000 embryos? Does the validation population include Asian populations?
- Consistency with Gold Standard: Is the concordance rate between AI interpretation results and traditional PGT-A (manual + software) above 95%? What is the detection rate for mosaicism?
- Correlation with Clinical Pregnancy Data: Are there published, peer-reviewed studies demonstrating a statistically significant improvement in clinical pregnancy rate or miscarriage rate per transfer cycle after using AI-assisted screening?
When is AI Genetics Screening Suitable?
Characteristics of Suitable Populations
- Female age ≥38 years, with high risk of embryonic aneuploidy, requiring priority screening for chromosomally normal embryos.
- History of recurrent implantation failure (≥3 failed transfers of good-quality embryos) or recurrent miscarriage (≥2 losses).
- One partner is a carrier of a balanced translocation, Robertsonian translocation, or other structural abnormality.
- Large number of embryos (≥8 blastocysts), requiring efficient ranking assistance to reduce selection difficulty.
Characteristics of Unsuitable Populations
- Very low ovarian reserve (AMH < 0.5 ng/mL), with expected oocyte retrieval ≤2. The incremental clinical value of AI screening is limited, and biopsy may affect embryo utilization.
- Those with a confirmed monogenic disease (e.g., thalassemia, hereditary cancer syndrome) requiring PGT-M; AI genetics screening cannot replace site-specific testing.
- Individuals highly sensitive to mosaic results, requiring absolute exclusion of any abnormal signals; note the potential for false positives with AI, necessitating manual review.
- Those who do not accept embryo biopsy or have ethical concerns about invasive procedures on embryos.
Actual Process and Timeline
Standard Process
- Ovarian Stimulation and Egg Retrieval: Standard controlled ovarian stimulation protocol, followed by in vitro fertilization after egg retrieval.
- Blastocyst Culture and AI Morphology Scoring: Culture to days 5-7, using the AI system for morphology and kinetic scoring of blastocysts, generating an initial quality ranking.
- Embryo Biopsy: Trophectoderm biopsy is performed on blastocysts meeting biopsy criteria (usually 4BB or higher), obtaining 5-10 cells.
- Whole Genome Amplification and NGS Sequencing: Biopsied cells are amplified and sequenced on a machine, generating raw genetic data.
- AI-Assisted Genetic Data Analysis: Sequencing data is input into the AI system, which outputs results such as chromosome copy number variations and mosaicism percentages.
- Manual Review and Genetic Counseling: Embryologists and reproductive physicians manually review the AI results, issue the final PGT-A report, and provide genetic counseling.
- Frozen Embryo Transfer or Further Screening: Based on the report, select transferable embryos and schedule a frozen-thawed embryo transfer cycle.
What to Prepare
- Complete Fertility Assessment Data: Including AMH, FSH, antral follicle count, thyroid function, semen analysis, etc.
- Chromosome Karyotype Analysis Report: Required from both partners to rule out structural abnormalities.
- Previous Reproductive History: Results of chromosome analysis of miscarried embryos (if available), transfer records, hysteroscopy reports, etc.
- Genetic Counseling Records: For carriers of monogenic diseases or chromosomal abnormalities, provide proof of genetic counseling.
How Long Does It Take
From the start of ovarian stimulation to obtaining the final PGT-A (including AI-assisted) report, it typically takes 4-6 weeks. This includes 5-7 days for blastocyst culture, and 7-14 days for genetic testing and AI analysis after biopsy. If a frozen embryo transfer protocol is used, an additional 1-2 menstrual cycles are needed for endometrial preparation.
Easily Overlooked Details
Technical Understanding Level
- AI Systems Require Continuous Updates: Algorithm models need regular retraining with new data, otherwise accuracy may decrease over time. When choosing a center, confirm whether its AI system has a regular update mechanism.
- Significant Differences Between AI System Brands: The AI systems used in Thailand mainly come from Europe, the US, Israel, and local development. Different systems have different algorithm architectures and validation data, so they cannot be generalized.
Practical Operation Level
- Biopsy Technical Quality Directly Affects AI Analysis: If the quality of biopsied cells is poor (e.g., cell lysis, uneven DNA amplification), even the best AI algorithm cannot produce accurate results.
- AI Results Must Include a Manual Review Step: Any AI system should be accompanied by a dual verification process by an embryologist/geneticist. AI reports lacking manual review should not be used as the sole basis for transfer.
Common Misconceptions and Pitfall Reminders
Misconception 1: AI Screening = No Need for PGT-A
AI screening is an auxiliary tool for PGT-A, not a replacement. Without PGT-A sequencing data, AI cannot independently perform genetic analysis. When choosing an AI system, it is essential to confirm that the center also has conventional PGT-A capabilities.
Misconception 2: AI Results are 100% Accurate
Any genetic screening has false positives and false negatives. AI systems may misjudge certain chromosomal abnormalities (e.g., inversion of chromosome 9, low-level mosaicism). Prenatal diagnostic methods such as ultrasound and NIPT should still be considered before transfer.
Misconception 3: AI Systems in Thailand are More Advanced Than Those Domestically
Some centers in Thailand have indeed introduced newer AI equipment, but leading domestic fertility centers (e.g., Peking University Third Hospital, CITIC Xiangya, Renji Hospital) also have their own technological expertise in AI-assisted embryo evaluation. Technological advancement needs to be assessed on a case-by-case basis, not simply judged by country.
Frequently Asked Questions
Can AI screening completely replace traditional PGT-A?
No. AI is an auxiliary tool in the data analysis stage and does not change the biological basis of PGT-A. Without embryo biopsy and NGS sequencing, AI cannot produce genetic results. The two have a collaborative upstream-downstream relationship, not a replacement relationship.
What is the difference between AI screening in Thailand and domestically?
The main differences are: ① Different technology sources; Thailand mostly uses European, American, or local algorithms, while some domestic centers use self-developed models. ② Different approval and regulatory systems; Thailand has relatively relaxed regulation of AI medical software, while domestically it must comply with medical device classification management regulations. ③ Different fee structures; AI screening in Thailand is usually charged as an additional service, while domestically it is often included in the PGT-A package.
Factors influencing the cost of AI screening
The additional cost for AI genetics screening in Thailand is typically between 30,000 and 80,000 Thai Baht (approximately 6,000-16,000 RMB), depending on the center's qualifications, the AI system brand, whether manual review is included, and the number of embryos. This fee does not include the base PGT-A cost (usually 100,000-200,000 Thai Baht).
Case Scenario Analysis
Scenario 1: Advanced Maternal Age with Recurrent Miscarriage
A 42-year-old patient with a history of 3 miscarriages; product of conception testing indicated trisomy 16. After traveling to Thailand for AI-assisted PGT-A screening, 6 blastocysts were obtained. AI scoring combined with traditional NGS results indicated 2 embryos were euploid. One embryo was transferred, resulting in a successful pregnancy, and prenatal diagnosis confirmed a normal karyotype. In this case, the main value of AI was rapid screening, shortening decision-making time.
Scenario 2: Risk of Missed Mosaicism
A 38-year-old patient; the AI system indicated one embryo as a low-level mosaic (approximately 15%), but the traditional software did not report an abnormality. Subsequent FISH verification confirmed true mosaicism. In this case, AI played a supplementary identification role, but subsequent confirmation via FISH or array-CGH was still required. This illustrates that AI results cannot be used as an independent decision-making basis.
Practitioner's Observation
As a reproductive physician, I have observed two trends in the development of AI genetics screening systems in Thailand: first, more and more centers are using AI as a point of technological differentiation; second, the lack of unified industry standards leads to uneven quality. For patients, rather than focusing on "whether AI is available," it is more important to focus on "whether the AI system has been independently validated," "whether the manual review process is robust," and "whether the laboratory has JCI or ISO certification." Technology is a tool; the core of clinical decision-making remains the physician's comprehensive judgment and the laboratory's quality control level.
End: Risk Reminder
