Artificial Intelligence in Obstetrics and Gynaecology

Quick Definition

• Machine Learning (ML) - algorithms that learn from data
• Deep Learning (DL) / Neural Networks - multilayered models for image analysis
• Natural Language Processing (NLP) - text/speech understanding
• Convolutional Neural Networks (CNNs) - specialized for image classification

FLOWCHART: AI in Obstetrics and Gynaecology

                    ┌─────────────────────────────┐
                    │   AI IN OBSTETRICS & GYN     │
                    └─────────────┬───────────────┘
                                  │
        ┌─────────────────────────┼──────────────────────────┐
        │                         │                          │
        ▼                         ▼                          ▼
┌───────────────┐      ┌──────────────────┐       ┌──────────────────┐
│  OBSTETRICS   │      │   GYNAECOLOGY    │       │  REPRODUCTIVE    │
│               │      │                  │       │   MEDICINE       │
└──────┬────────┘      └────────┬─────────┘       └───────┬──────────┘
       │                        │                          │
       ▼                        ▼                          ▼
┌──────────────┐     ┌────────────────────┐    ┌────────────────────┐
│ 1. Prenatal  │     │ 4. Gynae Oncology  │    │ 7. IVF / ART       │
│    Imaging   │     │                    │    │                    │
│              │     │ • Cervical cancer  │    │ • Embryo grading   │
│ • Fetal      │     │   screening (AI    │    │   (AI selects best │
│   biometry   │     │   Pap/colposcopy)  │    │   embryo to        │
│   (auto-     │     │ • Ovarian mass     │    │   transfer)        │
│   measurement│     │   characterization │    │ • Sperm analysis   │
│   of BPD,    │     │   (IOTA+AI)        │    │ • Ovarian reserve  │
│   FL, AC)    │     │ • Endometrial      │    │   prediction       │
│              │     │   cancer staging   │    │                    │
│ • Anomaly    │     │                    │    └────────────────────┘
│   detection  │     └────────────────────┘
│   (neural    │
│   tube,      │     ┌────────────────────┐    ┌────────────────────┐
│   cardiac)   │     │ 5. Endometriosis   │    │ 8. Robotic         │
│              │     │                    │    │    Surgery         │
│ • Placenta   │     │ • Non-invasive     │    │                    │
│   localiz-   │     │   diagnosis using  │    │ • AI-guided        │
│   ation &    │     │   MRI+AI pattern   │    │   instruments      │
│   grading    │     │   recognition      │    │ • Real-time        │
└──────┬───────┘     └────────────────────┘    │   tissue           │
       │                                        │   identification   │
       ▼             ┌────────────────────┐    └────────────────────┘
┌──────────────┐     │ 6. Urogynaecology  │
│ 2. Fetal     │     │                    │
│  Monitoring  │     │ • Pelvic floor     │    ┌────────────────────┐
│  (CTG/KTG)   │     │   imaging (MRI/    │    │ 9. Telemedicine    │
│              │     │   ultrasound +AI)  │    │  & Remote Care     │
│ • AI-CTG     │     │ • Incontinence     │    │                    │
│   interpre-  │     │   classification   │    │ • Wearables for    │
│   tation     │     └────────────────────┘    │   fetal heart      │
│ • Reduces    │                               │   rate monitoring  │
│   false +ve  │                               │ • Gestational      │
│ • Predicts   │                               │   diabetes         │
│   fetal      │                               │   monitoring       │
│   distress   │                               └────────────────────┘
└──────┬───────┘
       │
       ▼
┌──────────────────┐
│ 3. Risk Prediction│
│   Models         │
│                  │
│ • Pre-eclampsia  │
│   (1st trimester │
│   MAP + uterine  │
│   artery RI +    │
│   PLGF + AI)     │
│ • Preterm birth  │
│ • GDM prediction │
│ • PPH risk       │
└──────────────────┘

Detailed Application - Domain by Domain

1. Prenatal Imaging and Ultrasound

• Auto-measure fetal biometry (BPD, FL, HC, AC) - removes operator variability, saves 30-50% scan time
• Detect anomalies - CNS defects (neural tube, ventriculomegaly) with ~96-97% accuracy; reduced false negatives by 97.5% in one study
• Guide probe position - AI tells the sonographer in real-time when the correct standard plane is achieved (used in SonoSCOUT, GE Voluson AI)
• 3D reconstruction of fetal face/heart from 2D sweeps

2. Fetal Heart Rate Monitoring (CTG / Cardiotocography)

CTG Trace Generated
       │
       ▼
AI Algorithm analyzes:
  - Baseline FHR
  - Variability patterns
  - Accelerations
  - Decelerations (types)
  - Contraction relationship
       │
       ▼
Classifies as:
  Normal / Suspicious / Pathological
       │
       ▼
Clinical Alert sent to OBstetrician
  (reduces inter-observer disagreement)
       │
       ▼
Decision: Continue labour / Expedite delivery

• AI-CTG reduces unnecessary caesarean sections by reducing false-positive "non-reassuring" interpretations
• Systems like Monica AN24, Oxford Medilog use AI for CTG analysis
• Can predict fetal acidosis (pH <7.1) with higher accuracy than visual interpretation alone

3. First Trimester Risk Prediction (Pre-eclampsia, Preterm Birth)

11-13+6 weeks scan
         │
         ▼
AI Model inputs:
  ├── Maternal demographics (age, BMI, race)
  ├── Mean Arterial Pressure (MAP)
  ├── Uterine Artery Pulsatility Index (UtAPI)
  ├── Serum PLGF (Placental Growth Factor)
  └── Uterine PAPP-A
         │
         ▼
Risk Algorithm (FMF Combined Test)
  (Bayes' theorem + machine learning)
         │
         ▼
OUTPUT:
  ├── LOW RISK → Routine antenatal care
  └── HIGH RISK (>1:100) → Aspirin 150mg at night
              → Refer MFM
              → Doppler surveillance

• Detection rate: ~75% for early-onset pre-eclampsia at 11-13 weeks (FMF model)
• AI improves prediction by integrating non-linear relationships between biomarkers

4. Gynaecological Oncology - AI Applications

a) Cervical Cancer Screening

Cervical Sample / Colposcopy image
           │
           ▼
AI (Deep Learning CNN) analyzes:
  - Cell morphology (Pap smear)
  - Acetowhite patterns (VIA/VILI)
  - Colposcopy image features
           │
           ▼
Classifies: Normal / LSIL / HSIL / Cancer
           │
           ▼
Triage: Routine recall / Early treatment / Biopsy

• WHO-supported AI tools for cervical screening in low-resource settings (e.g., PAVE study - WHO/NCI VIA AI project)
• FDA-cleared: Hologic ThinPrep Imaging System uses AI for Pap smear analysis
• Reduces workload on cytopathologists by 50-70%

b) Ovarian Mass Classification

Pelvic Ultrasound
       │
       ▼
IOTA (International Ovarian
Tumor Analysis) Rules + AI
       │
       ├── Simple Rules (SR)
       ├── ADNEX Model
       └── AI augmentation
              │
              ▼
Classification:
  Benign / Low-risk / High-risk malignancy
              │
              ▼
Management:
  Conservative / MDT / Surgery

• AI-augmented IOTA-ADNEX model achieves AUC >0.95 for malignancy detection
• Reduces unnecessary surgical referrals

5. In-vitro Fertilization (IVF) and ART

Oocyte retrieval
       │
       ▼
Fertilization
       │
       ▼
Embryo Culture (Day 3 / Day 5)
       │
       ▼
AI-Timelapse Imaging (EmbryoScope)
  Analyzes:
    - Cell division timing (t2, t3, t5, tB)
    - Morphokinetics
    - Blastocyst expansion grade
       │
       ▼
AI Score: ViQi / KIDScore / iDAScore
       │
       ├── High Score → Priority transfer
       └── Low Score → Freeze / Discard

• AI embryo selection improves live birth rates vs. conventional morphological grading
• iDAScore (Vitrolife) and LifeWhisperer (Presagen) are FDA/CE-cleared AI embryo graders
• AI sperm analysis (ISAS, SCA): motility, morphology, DNA fragmentation - more objective than manual count

6. Intrapartum Care

7. Robotic Surgery - AI Integration

• da Vinci system + AI: real-time tissue identification, haptic feedback simulation
• AI assists in identifying ureters, vessels during laparoscopic hysterectomy (prevents inadvertent injury)
• Augmented reality overlay - visualises structures during myomectomy, endometriosis excision
• Reduces operative time and conversion rates in complex cases

8. Telemedicine and Remote Monitoring

• Wearable CTG devices (e.g., Monica AN24 patch) transmit FHR to AI platform - allows home monitoring of high-risk pregnancies
• AI chatbots for antenatal triage (FLORENCE, WHO)
• Gestational diabetes apps - continuous glucose monitoring + AI dietary advice
• Blood pressure AI apps - detect hypertension in pregnancy remotely

9. Ethical and Practical Challenges (Important for DNB!)

CHALLENGES OF AI IN OB/GYN
           │
     ┌─────┴──────┐
     │            │
  Clinical      Ethical
     │            │
  • Algorithm  • Data privacy
    bias (if     (patient
    trained      consent)
    on non-    • Algorithmic
    diverse      bias (under-
    data)        representation
  • Black-box    of minorities)
    problem    • Liability
  • Integration  (who is
    into EHR     responsible
    workflow     if AI errs?)
  • Training   • Equity of
    clinicians   access
  • Regulation
    (FDA 510k /
    CE marking)

10. Regulatory Framework

• FDA (USA): AI/ML-based Software as Medical Device (SaMD) - 510(k) or De Novo pathway
• CE marking (Europe): Medical Device Regulation (MDR 2017/745)
• India: CDSCO regulates AI-based medical devices under Medical Devices Rules 2017 (amended 2020)
• AI tools are classified as Class II or III depending on risk to patient

Summary Table for Quick Revision

Model DNB Answer Format (10 Marks)

• Ramirez Zegarra R & Ghi T. AI in fetal ultrasound. Ultrasound Obstet Gynecol 2023 [PMID 36436205]
• Horgan R et al. AI in obstetric ultrasound - scoping review. Prenat Diagn 2023 [PMID 37503802]
• Jiang VS & Bormann CL. AI in IVF laboratory. Fertil Steril 2023 [PMID 37211062]
• Moro F et al. AI in gynaecologic oncology ultrasound - systematic review. Int J Cancer 2024 [PMID 38989809]
• Wu T et al. AI in cervical cancer screening. Cancer Biol Med 2024 [PMID 39297572]
• PMC 12303583. AI in Obstetrics and Gynaecology - comprehensive review 2025