Predicting Cerebral Palsy in Infants With White Matter Injury Using MRI

Cerebral palsy (CP) is a common group of movement disorders that often results in disability in children. In the context of CP, the importance of early diagnosis is crucial, but current diagnostic modalities often identify cases after the age of 2 years. After initial screening of infants at high risk for CP by behavioral scoring, magnetic resonance imaging (MRI) forms an integral part of the comprehensive evaluation. The training of conventional model of CP risk prediction requires a large investment of time and financial resources. The average sensitivity rate drops to 90%. Up to now, deep learning technology has been widely used in tasks related to image-based disease classification and has shown excellent performance.

Periventricular white matter injury (PVWMI) accounts for the largest proportion of various types of brain injuries in cerebral palsy, and the types of brain injuries in cerebral palsy are rich and complex, posing difficulties and challenges to deep learning models. Therefore, this study focuses on PVWMI, the most common type of cerebral palsy, and uses conventional MRI to develop a deep learning prediction model for CP in infants aged 6 months to 2 years old.

Inclusion criteria

• Infants and children at high risk of periventricular white matter injury (PVWMI) (gestational age <35 weeks, birth weight <2.6 kg, forceps-assisted delivery/fetal head attraction, Apgar score <7, hypoglycaemia, sepsis, electrolyte disturbances, premature rupture of membranes);
• Those who underwent MRI at 6 months of age-2 years, including at least T1WI and T2WI sequences;
• Upon follow-up, the patient's clinical diagnosis: cerebral palsy, other diagnoses that did not develop into cerebral palsy, or inability to confirm the diagnosis).

Exclusion criteria

• Incomplete MRI images or unreadable images due to motion artefacts;
• Incomplete neurobehavioural assessment data (including: gross motor function).