Deep Learning-Based Craniosynostosis Classification via Suture Segmentation and Mask-Weighted Classification
Early diagnosis of craniosynostosis is critical to preventing neurological complications, yet skull X-ray interpretation remains subjective and existing models often rely on secondary deformations rather than primary pat
The authors studied a deep learning framework designed to classify craniosynostosis (CSO) using skull X-rays. This approach matters in metabolic and neurodevelopmental research contexts where early detection of skeletal anomalies can prevent severe neurological sequelae. Traditional diagnostic methods often depend on subjective interpretation or secondary signs, which may delay treatment.
To address these limitations, the researchers developed an Integrated Suture Segmentation and Classification Pipeline. They constructed a balanced dataset comprising 1,088 skull X-ray images from 368 unique patients. A segmentation model was trained to identify coronal, sagittal, and lambdoid sutures, followed by a classification stage that utilized predicted suture probability maps as weights.
Experimental results demonstrated that the proposed method with a DenseNet-161 backbone achieved an image-level Accuracy of 0.925 and an Area Under the Receiver Operating Characteristic curve (AUROC) of 0.980. Exam-level diagnosis via multi-view aggregation significantly improved performance, yielding an Accuracy of 0.941 and an AUROC of 0.994. Gradient-weighted Class Activation Mapping (Grad-CAM) analysis confirmed that the model's attention is primarily directed toward specific suture lines rather than global skull shape.
This study presents a clinically interpretable deep learning framework, highlighting its potential as a robust computer-aided referral decision support tool for primary care settings. The findings suggest that prioritizing anatomical features over secondary deformations can enhance diagnostic accuracy while minimizing the need for unnecessary radiation-intensive CT scans. Researchers note that this method facilitates timely specialist assessment.
It is important to frame these results within the context of research-use only applications. While the study demonstrates high performance, further validation in diverse clinical populations is required before widespread adoption. The authors emphasize that this work supports diagnostic decision-making rather than providing standalone medical advice or replacing clinical judgment.