Image processing and machine learning for telehealth craniosynostosis screening in newborns

Area of Science:

• Medical Imaging
• Pediatric Diagnostics
• Computational Biology

Background:

• Craniosynostosis is a condition where skull sutures fuse prematurely in infants.
• Accurate and timely diagnosis is crucial for proper neurodevelopment.
• Current diagnostic methods can be resource-intensive.

Purpose of the Study:

• To evaluate the accuracy of a novel telehealth-compatible software for diagnosing craniosynostosis in newborns.
• To compare the software's diagnostic performance against craniofacial specialists.
• To assess the software's agreement with gold-standard craniometric measurements.

Main Methods:

• A dataset of 351 newborn cranial images was analyzed.
• Software classification accuracy, sensitivity, and specificity were calculated.
• Agreement with specialist diagnoses and optical craniometrics was assessed using intraclass correlation coefficients.

Main Results:

• The software achieved 93.3% accuracy in classifying cranial shapes.
• Sensitivity was 92.0% and specificity was 94.3%.
• High agreement was found with optical craniometrics for cephalic index (0.95) and cranial vault asymmetry index (0.67).

Conclusions:

• The software demonstrates high accuracy for remote screening of nonsyndromic craniosynostosis in newborns.
• It can serve as a substitute for traditional craniometric assessments.
• Future development may enable mobile-based screening in primary care settings via telemedicine.