Prenatal neurosonography plays a vital role in evaluating central nervous system development. However, detecting fetal cranial abnormalities remains operator-dependent and technically challenging. A multicentre, self-crossover randomised controlled trial investigated the Prenatal Ultrasound Diagnosis Artificial Intelligence Conduct System (PAICS). This innovative platform assists sonographers in detecting fetal intracranial malformations during prenatal screening. By assessing mid-level practitioners across high-volume hospitals, the study provides valuable real-world evidence regarding clinical artificial intelligence integration.

Trial Design and the PAICS Technology

The multicentre trial enrolled 1,584 singleton pregnancies between 11 and 32 weeks of gestation carrying high risk for fetal malformations. Researchers conducted the investigation across five tertiary centres in China, engaging sonographers who possessed three to eight years of scanning experience. This cohort represents the core ultrasound workforce in routine clinical practice.

Investigators applied a robust self-crossover randomized design. Participants were assigned in a 1:1 ratio to two diagnostic sequences: independent real-time scanning followed by PAICS-assisted offline review, or PAICS-assisted real-time diagnosis followed by independent offline review. The protocol incorporated a four-week washout period between evaluations to prevent diagnostic recall bias.

Allocation remained strictly concealed throughout the trial. An independent expert panel reviewed full video recordings to establish the reference standard. Sonographers were masked to anomaly status, while outcome assessors and expert reviewers remained masked to sequence allocation. The trial established primary superiority endpoints for sensitivity and non-inferiority margins for specificity across ten distinct intracranial malformations.

Diagnostic Superiority for Fetal Intracranial Malformations

Completed between September 2022 and November 2023, the trial demonstrated that PAICS-assisted diagnosis significantly enhanced sensitivity for detecting fetal intracranial malformations. In the fetal-based analysis, sensitivity increased by 8.7 percentage points compared to unassisted scanning. Moreover, this improvement occurred without sacrificing specificity, successfully meeting the pre-specified 5% non-inferiority margin.

The algorithmic assistance proved particularly effective for identifying structural abnormalities, including ventriculomegaly, holoprosencephaly, and corpus callosum dysgenesis. Because PAICS tracks standard axial intracranial planes continuously, it flags abnormal contours and missing structures in real time. Consequently, mid-level sonographers identified subtle intracranial defects that might otherwise escape human visual detection.

Additionally, researchers analyzed whether the tool impacted anomalies outside its primary diagnostic scope. Sonographers maintained high accuracy for non-target conditions during assisted examinations. The software prompts did not distract clinicians or induce diagnostic tunnel vision. Instead, the structured AI workflow encouraged meticulous anatomical surveys, allowing practitioners to detect diverse anomalies with heightened confidence.

Real-Time Scanning Versus Offline Review Modes

The trial design compared real-time clinical guidance against post-acquisition offline image review. The investigators discovered that real-time AI assistance provided superior benefits during live probe manipulation. Because ultrasound interpretation depends directly on obtaining correct anatomical planes, real-time feedback helps sonographers optimize transventricular and transthalamic views instantaneously.

When sonographers operated in real-time assisted mode, the software highlighted regions of interest directly on the monitor. Therefore, clinicians could pause, adjust probe alignment, and re-examine suspicious cranial landmarks immediately. This active guidance prevents missed diagnoses that might otherwise necessitate inconvenient patient recalls after delayed offline audits.

Nevertheless, offline review modes also demonstrated substantial diagnostic value. When practitioners reviewed stored video sequences with AI assistance, sensitivity remained significantly higher than unassisted reads. Consequently, medical centres can deploy this computational tool either as a live scanning co-pilot or as an automated quality-assurance tool during reporting. This flexibility enables institutions to integrate artificial intelligence seamlessly into existing radiology workflows.

Clinical Relevance for Indian Prenatal Healthcare

These trial findings carry significant relevance for maternal-fetal healthcare across India. Obstetric ultrasound practices in India exhibit wide variation in sonographer training, equipment sophistication, and patient workload. While premier fetal medicine centres offer outstanding diagnostic precision, district hospitals and rural clinics often lack specialized neurosonographers. Therefore, deploying validated artificial intelligence could bridge performance disparities between primary and tertiary centres.

Timely detection of congenital central nervous system malformations is especially critical under the Indian legal framework. The Medical Termination of Pregnancy Amendment Act permits pregnancy termination up to 24 weeks for severe fetal anomalies, subject to Medical Board authorization. Consequently, detecting severe intracranial lesions during the 18 to 22-week anomaly scan provides families with crucial reproductive choices.

However, clinical implementation in India requires stringent regulatory compliance. Diagnostic software must adhere strictly to the Pre-Conception and Pre-Natal Diagnostic Techniques Act, prohibiting any feature capable of fetal sex determination. Furthermore, Indian clinicians must treat algorithmic findings as clinical decision support rather than final diagnoses, ensuring physician oversight and personalized counseling.

Implementation Challenges and Future Perspectives

Despite these encouraging results, several implementation barriers require thorough evaluation prior to routine clinical adoption. Algorithmic generalizability represents a primary concern when deploying systems across ultrasound machines from diverse vendors. Factors such as acoustic shadowing, maternal obesity, and transducer variations can alter image resolution and diminish algorithmic sensitivity. Therefore, extensive multi-vendor validation in diverse demographic populations remains imperative.

Additionally, medical training programs must address the potential risk of clinical automation bias. If junior sonographers rely excessively on automated alerts, they might overlook rare developmental syndromes residing outside algorithm training databases. Medical educators should incorporate artificial intelligence literacy into obstetrics and radiology residency programs, training practitioners to cross-examine digital alerts critically.

Finally, healthcare administrators must evaluate computational costs and infrastructure demands in public hospital networks. Future prospective studies should evaluate whether real-time AI guidance shortens scan times, reduces unnecessary referral burdens, and improves postnatal pediatric outcomes. Combining expert clinical judgment with advanced machine learning offers a balanced approach to elevating prenatal neurological screening standards.

Frequently Asked Questions

How does AI assistance alter the detection rate of congenital brain defects during pregnancy?

Artificial intelligence noticeably increases diagnostic sensitivity for congenital brain defects without reducing specificity. Consequently, sonographers can detect subtle structural abnormalities that human eyes might overlook during standard examinations. The automated system identifies missing structures, ventricular enlargement, and midline shifts across standard cranial planes. Therefore, integrating such algorithms allows mid-level practitioners to achieve diagnostic accuracy comparable to seasoned fetal medicine experts, ultimately improving prenatal counseling and post-delivery neurosurgical planning.

Can AI algorithms identify intracranial anomalies that lie outside their original training dataset?

Current deep learning models primarily demonstrate superior detection for the specific lesions included in their original training algorithms. However, clinical evidence shows that AI assistance also provides modest collateral benefits for identifying anomalies beyond its direct scope. Because the system enforces rigorous anatomical plane acquisition, sonographers inspect adjacent structures more thoroughly. Clinicians must remember that AI cannot replace comprehensive diagnostic acumen when evaluating rare or atypical intracranial malformations.

What regulatory considerations apply to deploying prenatal ultrasound AI software in Indian healthcare?

Deploying prenatal diagnostic software in India requires strict adherence to the Pre-Conception and Pre-Natal Diagnostic Techniques Act, known as the PCPNDT Act. Therefore, developers and clinics must ensure that automated systems strictly avoid gender determination capabilities. In addition, software platforms must obtain validation from the Central Drugs Standard Control Organisation before routine commercial use. Clinicians must maintain full legal responsibility for final diagnostic reports, treating artificial intelligence solely as a secondary clinical decision support tool.

Disclaimer: This content is for informational and educational purposes only. It is not intended to substitute for professional medical advice, diagnosis, or treatment. Always seek the advice of your physician or another qualified healthcare provider with any questions you may have regarding a medical condition or treatment. Refer to the latest local and national guidelines for clinical practice.

References

1. Lei T et al. Evaluating AI-assisted detection of fetal intracranial malformations in prenatal ultrasound practice: a multicentre, self-crossover, randomised controlled trial in China. Lancet Digit Health. 2026 Sep 10. doi: undefined. PMID: 42722598.
2. Lei T, Xie B, Wang N, et al. Use of real-time artificial intelligence in detection of abnormal image patterns in standard sonographic reference planes in screening for fetal intracranial malformations. Ultrasound Obstet Gynecol. 2022;59(5):668-677.
3. Salomon LJ, Al-Memar M, Bault JP, et al. Multicenter Evaluation of an Artificial Intelligence System for Automatic Recognition of Fetal Ultrasound Findings Suggestive of Congenital Malformations. Obstet Gynecol. 2026;147(2):245-254.