Comprehensive Summary
This study, conducted by Baptista et al., aimed to investigate the ability of hippocampal sclerosis detection in temporal lobe epilepsy through a comprehensive review of literature regarding radiomics and AI research in HS detection. This study utilized six research studies published up to May 2024 on Embase, Web of Science, and PubMed/Medline using PRISMA-DTA guidelines. STATA 14 was used for statistical analysis, sensitivity and specificity were analyzed with a bivariate model, and variation due to heterogeneity was analyzed with I2. The results showed that the support vector machine had the highest sensitivity and specificity, 0.92 and 0.95, then convolutional neural networks and logistic regression. The combination of AI and radiomics showed a lower sensitivity and specificity, 0.88 and 0.9, than AI alone, which had a sensitivity of 0.92 and specificity of 0.93. Overall, AI-based models had a combined sensitivity and specificity of 0.91 and 0.9, respectively. These results provide promising support for AI usage in accurate HS detection.
Outcomes and Implications
The complexity and ambiguity of the models are a drawback and prevent effective clinical applicability. However, if utilized correctly and efficiently, AI tools can lead to improved accuracy in diagnosis and determination of more effective treatment for patients.