Artificial Intelligence–Based Radiomic Model in Craniopharyngiomas: A Systematic Review and Meta-Analysis on Diagnosis, Segmentation, and Classification

Abstract

Background

Craniopharyngiomas (CPs) are rare, benign brain tumors originating from Rathke's pouch remnants, typically located in the sellar/parasellar region. Accurate differentiation is crucial due to varying prognoses, with adamantinomatous CPs having higher recurrence and worse outcomes. Magnetic resonance imaging struggles with overlapping features, complicating diagnosis. This study evaluates the role of artificial intelligence (AI) in diagnosing, segmenting, and classifying CPs, emphasizing its potential to improve clinical decision-making, particularly for radiologists and neurosurgeons.

Methods

This systematic review and meta-analysis assess AI applications in diagnosing, segmenting, and classifying on CP patients. A comprehensive search was conducted across PubMed, Scopus, Embase, and Web of Science for studies employing AI models in patients with CP. Performance metrics such as sensitivity, specificity, accuracy, and area under the curve were extracted and synthesized.

Results

Eleven studies involving 1916 patients were included in the analysis. The pooled results revealed a sensitivity of 0.740 (95% confidence interval [CI]: 0.673–0.808), specificity of 0.813 (95% CI: 0.729–0.898), and accuracy of 0.746 (95% CI: 0.679–0.813). The area under the curve for diagnosis was 0.793 (95% CI: 0.719–0.866), and for classification, it was 0.899 (95% CI: 0.846–0.951). The sensitivity for segmentation was found to be 0.755 (95% CI: 0.704–0.805).

Conclusions

AI-based models show strong potential in enhancing the diagnostic accuracy and clinical decision-making process for CPs. These findings support the use of AI tools for more reliable preoperative assessment, leading to better treatment planning and patient outcomes. Further research with larger datasets is needed to optimize and validate AI applications in clinical practice.