Precision psychiatry in schizophrenia: a meta-analysis of radiomics-based advanced neuroimaging for diagnostic and treatment-response prediction

Reliable neuroimaging biomarkers for the diagnosis of schizophrenia and the prediction of treatment response remain limited, particularly at the individual level. Radiomics converts neuroimaging data into high-dimensional quantitative data that may capture subtle disease-related patterns not detectable by conventional analyses. This study aimed to meta-analyze the diagnostic and treatment-response prediction performance of radiomics-based neuroimaging in schizophrenia.

A systematic search of PubMed, EMBASE, IEEE Xplore, and Scopus was conducted until 09/17/2026, following PRISMA-DTA guidelines and a pre-registered protocol (PROSPERO/ CRD420261286782). Eligible studies applied radiomics combined with machine learning/quantitative neuroimaging data (MRI/PET) for schizophrenia spectrum diagnosis/treatment response classification. A random-effects meta-analysis was performed to pool the area under the receiver operating characteristic (ROC) curve (AUC).

Hierarchical summary ROC (HSROC) models were used to estimate pooled sensitivity and specificity. Risk of bias and reporting quality were assessed using QUADAS-2 and METRICS. Fourteen studies were included.

For diagnostic discrimination, radiomics-based models showed good performance (AUC = 0.81, 95% CI [0.73; 0.88], I2 = 84.2%,k = 8). The HSROC model yielded a pooled mean sensitivity = 0.736, 95% CI [0.666; 0.795], and specificity = 0.735, 95% CI[0.668; 0.793]. Radiomics also showed good performance in differentiating treatment response, including longitudinal prediction (AUC = 0.80) and cross-sectional discrimination (AUC = 0.88).

Although substantial methodological heterogeneity among studies emerged, overall study quality was moderate/good. Radiomics-based neuroimaging demonstrated good diagnostic and treatment-response discrimination in schizophrenia, supporting their potential complementary function for precision psychiatry. However, limited external validation may currently constrain clinical translation, since less rigorous validation procedures were associated with more optimistic performances, as displayed by our analyses.

Future studies should prioritize standardized radiomics workflows, prospective designs, integration with clinical markers, and rigorous validation. Azienda Sanitaria Locale Napoli 3 Sud, U.O.S.M. 55-57 Torre del Greco - Ercolano, Torre del Greco, Italy Department of Advanced Medical and Surgical Sciences, University of Campania “Luigi Vanvitelli”, Naples, Italy Giancarlo Fusco, Lorenzo Ugga & Mario Cirillo Department of Psychiatry, University of Campania “Luigi Vanvitelli”, Naples, Italy Claudia Toni, Gaia Sampogna & Andrea Fiorillo The authors declare no competing interests. Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.