Recognizing Pediatric Tuberous Sclerosis Complex Based on Multi-Contrast MRI and Deep Weighted Fusion Network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pediatric Neurology

Background:

• Multi-contrast magnetic resonance imaging (MRI) is crucial for diagnosing tuberous sclerosis complex (TSC) in children.
• Accurate identification of TSC lesions is essential for timely clinical management.

Purpose of the Study:

• To develop and evaluate a deep convolutional neural network (CNN) for diagnosing pediatric TSC using multi-contrast MRI.
• To introduce a novel synthesized MRI modality (FLAIR3) for improved lesion contrast.

Main Methods:

• A deep weighted fusion network (DWF-net) employing a late fusion strategy was developed.
• A new synthesis modality, FLAIR3, was created by combining T2W and FLAIR MRI sequences.
• The DWF-net was trained and validated on a dataset of 680 children (349 TSC, 331 healthy).

Main Results:

• The FLAIR3 modality effectively enhanced the visibility of TSC lesions compared to standard sequences.
• The DWF-net achieved a high diagnostic performance with an AUC of 0.998 and an accuracy of 0.985.
• The proposed method significantly outperformed previous diagnostic approaches.

Conclusions:

• The novel FLAIR3 synthesis and DWF-net offer a promising advancement in pediatric TSC diagnosis.
• This AI-driven approach has the potential to serve as a reliable computer-aided diagnostic tool for radiologists.
• Improved lesion detection and classification can lead to better clinical outcomes for children with TSC.