Deep Learning-based approach for screening neonatal cerebral lesions on ultrasound images in China: a stepwise, multicenter, early-stage clinical validation study

Timely and accurate diagnosis of severe neonatal cerebral lesions is critical for preventing long-term neurological damage and addressing life-threatening conditions. Cranial ultrasound (CUS) is the primary screening tool, but the process is time-consuming and reliant on operator proficiency, leading to variability in both image quality and diagnostic outcomes across different radiologists. While artificial intelligence (AI) has made significant strides in many areas of healthcare, its application in CUS screening remains limited. In this study, we developed and evaluated a Neonatal Cerebral Lesions Screening system (NCLS) capable of automatically extracting standard views from CUS videos and identifying cases with severe cerebral lesions. The system was trained and validated using a dataset of 8,757 neonatal CUS images. It demonstrated strong performance, achieving an area under the curve (AUC) of 0.982 and 0.969, with sensitivities of 0.875 and 0.885 on internal and external validation sets, respectively. Furthermore, the NCLS outperformed junior radiologists (with 1–2 years of CUS diagnostic experience) and performed comparably to mid-level radiologists (with 3–7 years of experience), with 55.11% faster examination efficiency.