GLAPAL-H: Global, Local, And Parts Aware Learner for Hydrocephalus Infection Diagnosis in Low-Field MRI

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pediatric Neurology

Background:

• Infant hydrocephalus diagnosis relies on costly and potentially harmful CT scans.
• Low-field MRI offers a safer, more accessible alternative but faces image quality challenges.
• Differentiating between post-infectious hydrocephalus (PIH) and non-post-infectious hydrocephalus (NPIH) is crucial for appropriate treatment.

Purpose of the Study:

• To develop and validate a novel method for classifying infant hydrocephalus using low-field MRI.
• To address image quality limitations inherent in low-field MRI for hydrocephalus diagnosis.
• To establish low-field MRI as a viable, cost-effective alternative to CT for pediatric hydrocephalus assessment.

Main Methods:

• Proposed GLAPAL-H (Global, Local, And Parts Aware Learner), a multi-task deep learning architecture.
• Implemented global and local feature extraction branches using CNNs.
• Developed three regularized loss functions for holistic, detailed, and soft segmentation mask learning.

Main Results:

• GLAPAL-H achieved superior performance in accuracy, interpretability, and generalizability for both two-class (PIH vs. NPIH) and three-class (PIH vs. NPIH vs. Healthy) tasks.
• Outperformed existing state-of-the-art methods, including CT-based approaches.
• Demonstrated robustness against variations in training data quantity and quality.

Conclusions:

• GLAPAL-H demonstrates the potential of low-field MRI for safer, low-cost pediatric hydrocephalus diagnosis.
• The method enhances deployability due to its robustness in varied imaging conditions.
• GLAPAL-H offers a promising advancement in diagnosing and managing infant hydrocephalus.