Comprehensive Summary
The authors leverage machine learning (ML) to identify which of the five major brain networks have the greatest predictive power for therapeutic outcomes in a transcranial magnetic stimulation (TMS)–based smoking-cessation therapy. The five large-scale brain networks commonly reported in addiction research are the default mode network (DMN), central executive network (CEN), dorsal attention network (DAN), salience network (SN), and reward network (RN). Specifically, average functional connectivity of these networks (during exposure to smoking cues as well as at rest) was used as input features. Using the statistical modeling software SPSS 29.0, the researchers constructed a simple neural network trained to predict clinical outcomes from pre-treatment, network-specific connectivity measures. Feature importance analysis revealed that connectivity of the SN during both cue exposure and rest was the strongest predictor of clinical outcomes. Interestingly, high SN connectivity during task-fMRI (smoking cues) predicted favorable outcomes, whereas low SN connectivity at rest predicted the same. These reverse predictive effects suggest that further investigation is needed to clarify how addictive behavior is encoded across networks. Overall, the findings highlight the SN as a key target for future TMS-based interventions in smoking cessation.
Outcomes and Implications
Feature importance is a technique that helps researchers understand how an ML model makes its decisions. While this becomes more challenging as model architectures grow more complex, the paper sets a useful precedent by applying ML methods to identify regions of interest for future research. Notably, the finding that the salience network (SN) was most predictive of clinical outcomes motivated the researchers to target nodes within that network to iteratively refine treatment. Overall, integrating ML models with fMRI data to guide neuromodulation therapies is a promising approach that leverages modern tools to optimize clinical outcomes.