Role of AI in Reproductive Medicine

Advances in AI are proportionately connected with more data flowing in reproductive medicine. Fertility treatment varies with individuals, with no one size fits all. With progress in assisted reproductive technology (ART) treatment like oocyte or embryo cryopreservation, PGD—preimplantation genetic testing, embryo selection techniques, clinical pregnancy rates have improved. The quality of embryo is the most crucial factor, and there are no methods available to judge the same. The embryo, egg or sperm selection methods have not yet been identified. Hence, it is difficult to understand the reason for failure or parameters to predict success. AI can optimize the treatment using huge data with complex diagnosis and therapeutic modalities for successful results with less financial burden.

ApplicationsOocyte Selection

Good quality oocytes give high success rate. Many strategies have been proposed to evaluate and select oocyte with best developmental potential, but still, it may not be able to give results. The best method of oocyte selection should be non-invasive, not expensive and capable of getting incorporated into embryology system with least impact. AI methods will help evaluate human oocyte with good developmental potential.

Embryo Selection

Embryologist uses visualization method (morphology—dynamic development) to select embryos or oocytes, and evaluation is subjective. There is more possibility of pre-eclampsia, multi-foetal pregnancy, maternal haemorrhage, if more than one embryo is transferred. Morphology of embryo remains the main factor for selection. AI has optimized culture condition of embryo, improving its survival and development. Grading and ranking embryos help predict decision making and AI developed morphokinetic model that can exclude embryos with lowest implantation potential.

Sperm Selection/Semen Quality

The computer-aided sperm analysis (CASA) is used for semen examination. Precise results are challenging, because of difficult evaluation of sperm morphology manually and no uniformity between laboratories. One-third of male factor fertility are idiopathic, and the current method of semen analysis cannot detect multiple male factors. There is link between semen quality and environmental factors. Prediction of chromosomal abnormality could be up to 95%, while taking into consideration height, testicular volume, follicular-stimulating hormone (FSH), luteinizing hormone (LH), testosterone and ejaculate volume [1]. AI techniques were used taking into consideration these factors, and it helped improve performance with avoidance of environmental factors with better results.

Predictive Model Creation

With AI, clinicians can create personalized treatment module of ART, predict optimal time of transfer and improve pregnancy outcome. AI incorporated accuracy of prediction is gradually improving after using previous in vitro fertilization (IVF)/intracytoplasmic sperm injection (ICSI) records. The crucial predictive factors being age of women, no of developed embryos, serum estradiol levels on day of human chorionic gonadotropin (HCG) administration. Supervised learning method has developed strategies for female age-related different embryo transfer.

AI with ML can dramatically promote reproductive medicine in near future, as there are ongoing studies on good non-invasive marker to improve implantation rate and efficacy of treatment. Limitations being retrospective small data from single source, no randomized control trials, ethical consideration and possibility of risk of offsprings.

ObstetricsFoetal Heart Monitoring/Pregnancy Surveillance

Foetal heart rate (FHR) pattern shows cardiac and neurological responses to haemodynamic changes. CTG (cardiotocography) has been able to reduce neonatal complications, but regular observations are required to take prompt/timely decisions. Systematic review did not show any improvement in neonatal outcome after ML interpretation because AI models that are used were based on human interpretation [4]. With advanced computer system and engineering theory, leading to automatic interpretation of CTG without human interpretation, it will be able to give accurate results [5]. It can be used to monitor high risk pregnancy, to identify foetus at distress.

Ultrasonography (USG)

Obstetrics USG measurements, though standardized can be challenging at times in case of obesity, acoustic shadow, speckled noise. Currently manual USG methods are slow, with possibility of subjective errors. Data storage of 2D USG images is not very useful. New technology is being improvised to improve the acquired image to standardize measurements. ML has been able to distinguish different foetal body parts, and semi-automatic programmes have been implemented for body measurement using AI algorithm after appropriate body part is selected by the sonologist. Automated programmes that are in service are foetal head measurements, congenital malformations, nuchal translucency, foetal heart, AFI and type of cervix that can be used by trained professionals. In late first trimester, difference in texture of placenta has been studied in hypertensive disorders of pregnancy (HDP) to predict possibility of abnormal outcome in late pregnancy before clinical presentation [6]. Training is required to get high quality image within appropriate time to be able to capture in image scanning procedure. Deep learning model is being developed by using motion, action and pupillary response of sonologist with safety issue information.

Foetal Echocardiography

Foetal echocardiography is useful for making diagnosis of congenital cardiac anomalies, monitoring foetal growth restriction (FGR) and twin-to-twin transfusion syndrome (TTTS). Performing the same is challenging due to small heart, faster heart rate, involuntary foetal movements, limited access and lack of expertise. Intelligent navigation method like ‘FINE’ has been developed to detect cardiac anomalies, and deep learning (DL) method has been devised to have essential view of heart with structures from all dimensions. Limitation being proper assessment of all images with clinical decision is difficult.

Preterm Labour

AI is being used for measurement of gestational age, amniotic fluid index (AFI), cervical length and prediction of preterm labour. Deep learning base programmes have been devised to measure angle of progression for suspected preterm labour.

International Society of Ultrasound in Obstetrics & Gynaecology has introduced ‘Artificial Intelligence in imaging’ session [4]. Apart from taking measurements, system is devised to use this data to make diagnosis for further treatment directions.

Magnetic Resonance Imaging (MRI)

AI is used in obstetrics for foetal brain and for diagnosis of placenta accreta/previa. With AI techniques, ventriculomegaly that has been diagnosed antenatally could predict of the possibility of further treatment after birth with great accuracy. AI-based MRI scans gave accurate diagnosis of placenta accreta/percreta and in TTTS, accurate information of volume and distribution of vessels over placenta.

Gynaecological Oncology

Gynaecological cancer prognosis depends upon International Federation of Obstetrics & Gynaecology (FIGO) classification. With advent of new radiological and bio markers, there is impact on the treatment. For endometrial cancer, p53/KRAS gene mutation values and extramural vascular invasion of pelvic tumours on radiology have stratification value [3]. Despite research, there is a challenge to treat cancer due to multi-factorial aetiology. Predicting response to neoadjuvant therapy on individual basis needs deeper understanding, and AI algorithms are being developed to tackle this.

Example would be Software text lab 2, for epithelial serous ovarian tumour, where apart from CA 125 levels, other factors like size, shape, texture, wavelength, intensity were considered, and prediction of prognosis was established with regards response to chemo and surgical therapy [3]. In patients with CIN (cervical intraepithelial neoplasia), after using data of (human papilloma virus) HPV DNA and colposcopy, ANN’s prediction of progression to carcinoma has specificity of 99% with sensitivity of 93%, which gives an opportunity for timely intervention for better patient care [3].

Personalized Medicine

Personalized medicine means use of combined knowledge of an individual like genetic, specific medical history to predict possibility of disease with its prognosis and response to treatment. It can guide patient management and forecast outcome. Example would be BRCA 1/BRCA 2 gene for breast cancer predictability, KRAS for endometrial cancer, WNT signal for ovarian—endometrial—cervical cancer prognosis on individual basis [3]. This has caused paradigm shift in medical management to preventive precision care that required detailed vast data. AI has managed to synthesize data in oncology with individual medicine.

Gynaecology

AI has been used for diagnosis of endometriosis and predict growth/behaviour of fibroid from imaging data. In urogynaecology, MRI images could help diagnosis/quantification of pelvic organ prolapse (POP), understand urodynamic study and subgroups of urinary incontinence (OAB–Overactive Bladder). However, post-therapy (surgical–non-surgical), improvement in quality of life could not be ascertained due to limited data/literature.

ChatGPT and AI

ChatGPT was produced by OpenAI in Nov 2022 [7]. ChatGPT has potential to impart preliminary information on most of the topics in Obstetrics and Gynaecology. ChatGPT has power of deep learning that can mimic human language. Chatbot has 2 components—versatile AI system and chat interface that enables to have interactive session through queries, followed by response, and emulating human conversation. ChatGPT can serve as ‘clinical assistant’ and help in research writing. It has ability to extract data from electronic medical records and literature search, providing guidance on formatting and writing style. Medical personnel can then review and edit to have precision on what is relevant. It can collaborate multiple reviewer’s opinion to have conclusive outcome that can speed up medical writing with accuracy.

ChatGPT can help in formulating differential diagnosis, explanation in simple language about emergency obstetric situation, early pregnancy diagnosis, peri-partum care, family planning and sexual health. Management after diagnosis is articulate and well informed. Though ChatGPT has been able to outperform humans, challenge is on accuracy and reliability so far information on various topic is concerned. Occasional lapses in understanding the content of question can lead to misinformation and misinterpretation (hallucination). Hence, it does not replace medical advice given by the clinician but that should not deter clinician by exploring more uses of ChatGPT.