Identification of Children's Tuberous Sclerosis Complex with Multiple-contrast MRI and 3D Convolutional Network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pediatric Neurology

Background:

• Tuberous Sclerosis Complex (TSC) diagnosis in children is challenging.
• Magnetic Resonance Imaging (MRI) is a key diagnostic tool for TSC.
• Improving lesion conspicuity and diagnostic accuracy is crucial.

Purpose of the Study:

• To develop a novel MRI modality (FLAIR3) by combining T2w and FLAIR sequences.
• To propose a deep learning approach using two 3D Convolutional Neural Networks (CNNs) with late fusion for TSC diagnosis.
• To evaluate the efficacy of FLAIR3 and the proposed diagnostic model in identifying TSC in children.

Main Methods:

• A new MRI sequence, FLAIR3, was created by merging T2w and FLAIR images.
• Two distinct 3D CNN models were employed.
• A late fusion strategy was utilized to combine the outputs of the two CNNs.
• The study included 520 children (260 healthy, 260 with TSC).

Main Results:

• FLAIR3 enhanced the visibility of TSC lesions and improved classification performance.
• The proposed late fusion deep learning model achieved state-of-the-art performance.
• The model demonstrated an Area Under the Curve (AUC) of 0.994 and an accuracy of 0.971.
• The method proved effective in distinguishing between healthy children and those with TSC.

Conclusions:

• The developed deep learning method serves as an effective computer-aided diagnostic tool for TSC in children.
• FLAIR3 offers a novel imaging modality for precise localization of TSC lesions.
• This non-invasive approach can significantly aid clinical radiologists in diagnosing TSC patients efficiently and reliably.