Author(s) :

Harshatej Simhadri.

Volume/Issue :

Abstract :

The Response Assessment in Neuro-Oncology (RANO) criteria are widely used for assessing treatment response in brain tumours; however, they have recognized limitations in differentiating true tumour progression from treatment-related imaging effects such as pseudo-progression. Radiomic analysis enables non-invasive evaluation of tumours by extracting numerous quantitative features from medical images, thereby revealing imaging characteristics that may not be detectable through standard visual interpretation. This systematic review evaluates existing evidence on machine learning-enhanced radiomic applications in neuro-oncology, specifically the prediction of treatment response, molecular marker characterization, and survival prognostication. The quality of the selected studies was assessed using the QUADAS-2 tool and the TRIPOD guidelines for prediction model studies. Data synthesis was conducted in accordance with PRISMA 2020 guidelines. There were 12,847 patients across 63 studies who met all the inclusion criteria. Multiparametric radiomic models incorporating shape, intensity, and texture features derived from T1-weighted, T2-weighted, and FLAIR sequences demonstrated higher reported performance metrics across all clinical applications. Nevertheless, there was substantial heterogeneity in the feature extraction protocols, the implementation of validation strategies, and approaches to model interpretation. Integrating machine learning techniques with radiomic feature analysis has become an advancing approach in precision neuro-oncology, often showing improved predictive accuracy compared with traditional evaluation strategies in multiple clinical settings. Successful clinical implementation will depend on standardized imaging acquisition practices, rigorous validation across multiple institutions, and the development of transparent and interpretable modelling approaches. Key future directions involve conducting prospective clinical studies, applying federated learning approaches, and incorporating these models into clinical decision-support platforms.