Foundation-Model-Boosted Multimodal Learning for fMRI-based Neuropathic Pain Drug Response Prediction

TL;DR

Neuropathic pain drug-response prediction from rs-fMRI is hindered by limited data. The authors introduce $FMM_{TC}$, a foundation-model-boosted multimodal framework that fuses Time Series ($\mathbf{X}_T$) and Functional Connectivity ($\mathbf{X}_C$) while leveraging external knowledge from a large pain-agnostic fMRI foundation model (BrainLM trained on UK Biobank data). The architecture uses a frozen BrainLM TS encoder and a learnable FC encoder with concatenation-based fusion, trained end-to-end under binary cross-entropy loss, achieving substantial improvements in cross-dataset prediction and drug-specific generalization (e.g., MCC gains of $\geq 14.71\%$ from external knowledge and $\geq 2.80\%$ from multimodal fusion). Interpretability via integrated gradients reveals dynamic modality reliance across datasets, supporting robust cross-domain adaptability. Overall, $FMM_{TC}$ holds promise for accelerating neuropathic pain clinical trials by improving responder stratification and reducing study costs.

Abstract

Neuropathic pain, affecting up to 10% of adults, remains difficult to treat due to limited therapeutic efficacy and tolerability. Although resting-state functional MRI (rs-fMRI) is a promising non-invasive measurement of brain biomarkers to predict drug response in therapeutic development, the complexity of fMRI demands machine learning models with substantial capacity. However, extreme data scarcity in neuropathic pain research limits the application of high-capacity models. To address the challenge of data scarcity, we propose FMM$_{TC}$, a Foundation-Model-boosted Multimodal learning framework for fMRI-based neuropathic pain drug response prediction, which leverages both internal multimodal information in pain-specific data and external knowledge from large pain-agnostic data. Specifically, to maximize the value of limited pain-specific data, FMM$_{TC}$ integrates complementary information from two rs-fMRI modalities: Time series and functional Connectivity. FMM$_{TC}$ is further boosted by an fMRI foundation model with its external knowledge from extensive pain-agnostic fMRI datasets enriching limited pain-specific information. Evaluations with an in-house dataset and a public dataset from OpenNeuro demonstrate FMM$_{TC}$'s superior representation ability, generalizability, and cross-dataset adaptability over existing unimodal fMRI models that only consider one of the rs-fMRI modalities. The ablation study validates the effectiveness of multimodal learning and foundation-model-powered external knowledge transfer in FMM$_{TC}$. An integrated gradient-based interpretation study explains how FMM$_{TC}$'s cross-dataset dynamic behaviors enhance its adaptability. In conclusion, FMM$_{TC}$ boosts clinical trials in neuropathic pain therapeutic development by accurately predicting drug responses to improve the participant stratification efficiency.

Figures (2)

• Figure 1: FMM$_{TC}$ pipeline. The raw rs-fMRI $\mathbf{X}_{raw}$ undergoes three processing steps to obtain time-series (TS) $\mathbf{X}_{T}$. Then, functional connectivity (FC) $\mathbf{X}_{C}$ is computed from $\mathbf{X}_{T}$. Next, $\mathbf{X}_{T}$ and $\mathbf{X}_{C}$ are processed by a time-series (TS) encoder $E_{T}$ (frozen fMRI foundation model) and a functional connectivity (FC) encoder $E_{C}$ (learnable CNN), respectively. Finally, features of two modalities $\mathbf{R}_{T}$ and $\mathbf{R}_{C}$ are fused to get the multimodal feature $\mathbf{R}_{TC}$ for the following prediction.
• Figure 2: (a) Multimodal feature $\mathbf{R}_{TC}$ from FMM$_{TC}$ outperforms unimodal features from other feature extractors with linear classifiers on drug-agnostic response prediction. (b) Feature importance via integrated gradients (IG) illustrates how FMM$_{TC}$'s cross-dataset dynamic behaviors enhance adaptability: FMM$_{TC}$ flexibly prioritizes the most effective modality in predictions. The light blue background is IG values for time-series feature $\mathbf{R}_{T}$ and the light red region is for functional connectivity feature $\mathbf{R}_{C}$.