Radiomics, machine learning, and deep learning for hippocampal sclerosis identification: a systematic review and...

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Hippocampal sclerosis (HS) is a key pathology in temporal lobe epilepsy (TLE), often leading to difficult-to-treat seizures.
• Conventional diagnostic tools like EEG and MRI have limitations in accurately identifying HS.
• Artificial intelligence (AI) and radiomics offer promising non-invasive methods to enhance diagnostic precision.

Purpose of the Study:

• To systematically review and synthesize current research on AI and radiomics for improving the detection of HS in TLE.
• To evaluate the diagnostic performance of AI-based models in identifying HS, a common cause of refractory epilepsy.

Main Methods:

• A systematic literature search was conducted across PubMed/Medline, Embase, and Web of Science until May 2024, adhering to PRISMA-DTA guidelines.
• Statistical analysis involved pooling sensitivity and specificity using a bivariate model to assess AI model performance.
• Heterogeneity was evaluated using the I2 statistic, with data analyzed using STATA 14.

Main Results:

• Six studies were included, demonstrating pooled sensitivity of 0.91 and specificity of 0.9 for AI models in HS detection (AUC=0.96).
• AI models alone achieved higher performance (sensitivity: 0.92, specificity: 0.93) compared to AI combined with radiomics (sensitivity: 0.88, specificity: 0.9).
• Support vector machine (SVM) algorithms exhibited the highest diagnostic accuracy (SEN: 0.92, SPE: 0.95), followed by CNNs and LR.

Conclusions:

• AI models, especially SVM, demonstrate significant accuracy for HS detection in TLE.
• AI alone appears more effective than its combination with radiomics for HS diagnosis.
• Integrating AI into diagnostic workflows can facilitate earlier detection and personalized epilepsy management, improving patient outcomes.