Classification of paediatric brain tumours by diffusion weighted imaging and machine learning.

Area of Science:

• Neuroimaging
• Radiology
• Pediatric Oncology

Background:

• Posterior fossa brain tumors are a significant cause of morbidity in children.
• Accurate and early diagnosis is crucial for effective treatment planning.
• Distinguishing between different types of posterior fossa tumors can be challenging with conventional imaging.

Purpose of the Study:

• To evaluate the ability of apparent diffusion coefficients (ADC) to differentiate between common pediatric posterior fossa brain tumors.
• To assess the multicenter applicability of ADC histogram analysis for tumor classification.
• To determine the diagnostic accuracy of ADC metrics in classifying Medulloblastomas, Pilocytic Astrocytomas, and Ependymomas.

Main Methods:

• Diffusion-weighted imaging was performed on 124 pediatric patients across 12 centers using 18 scanners.
• Apparent diffusion coefficient (ADC) maps were generated, and histogram data was extracted from tumor regions of interest.
• Machine learning classifiers (Naïve Bayes, Random Forest) were trained using histogram metrics for classification, with accuracy assessed by tenfold cross-validation.

Main Results:

• Mean ADC values significantly differed between tumor types (ANOVA P < 0.001).
• A mean ADC cutoff of 0.984 × 10⁻³ mm²/s distinguished Ependymomas from Medulloblastomas with 80.8% sensitivity and 80.0% specificity.
• ADC histogram metrics achieved high classification accuracies: 85% with Naïve Bayes and 84% with Random Forest.

Conclusions:

• Apparent diffusion coefficient histogram analysis is a reliable method for classifying common pediatric posterior fossa brain tumors on a multicenter basis.
• ADC metrics offer a non-invasive approach to improve diagnostic accuracy and potentially guide treatment decisions.
• This technique holds promise for enhancing the diagnostic workflow in pediatric neuro-oncology.