Cost-Effectiveness and Methodological Analysis of fMRI-guided Transcranial Magnetic Stimulation Biomarker Studies

Abstract

The field of precision psychiatry relies on biomarkers to diagnose disease and predict treatment success (Cappon and Pascual-Leone 2024). Transcranial magnetic stimulation (TMS) can be used to treat psychiatric symptoms by inducing neural activity in a targeted region, and modulating activity in downstream circuits. TMS targeting using functional magnetic resonance imaging (fMRI) enables precise spatial targeting and mapping of each patient’s brain connectivity. Associations between baseline functional connectivity and symptom improvement have been proposed as candidate predictive biomarkers, as they may allow physicians to anticipate treatment response. However, progress in this field remains limited by challenges in replication and validation, driven in part by heterogeneity in how brain connectivity is processed and analyzed. In this work, we investigate methodological heterogeneity in image processing parameters, including differences in regression, frequency filtering, and spatial smoothing, as well as TMS modeling choices. Using data from the Symptom-Specific targets for TMS (SSTMS) clinical trial (n = 35) (Taylor et al. 2026), connectivity between stimulation sites and anxiosomatic and dysphoric circuits was computed and related to clinical outcomes (BAI and BDI). In parallel, using the Three-D dataset (n = 300) (Blumberger et al. 2018), we use a downsampling with replacement approach across increasing sample sizes to model variance, precision, and proportion of statistically significant results. A Value of Information cost-effectiveness interface was developed to inform sample size selection based on cost and expected gains in precision and proportion of statistically significant results: https://cost-effectiveness-dashboard.onrender.com. Our results show that preprocessing choices, particularly whole-brain regression (Li et al. 2019) and filtering order, significantly influence connectivity estimates and their influence on biomarker assessment. In contrast, spatial smoothing and white matter regression had minimal impact on overall conclusions. Although we did not propose a standardized methodological protocol, these findings highlight the need for further systematic evaluation of preprocessing pipelines. Overall, this work emphasizes the importance of methodological choices in fMRI-guided TMS studies and provides practical tools for improving reproducibility and optimizing experimental design.