313 Mapping common and distinct neurocircuitry: Information-theoretic integration of asymmetric CBMA datasets

Objectives/Goals: To develop an information-theoretic alternative to ALE that integrates uneven datasets using log-likelihood ratios (LLRs) under noise versus signal world models. This approach aims to improve reproducibility, signal-to-noise ratio, and computational scalability in neuroimaging meta-analysis. Methods/Study Population: We used both real and simulated coordinate-based datasets to build an information-theoretic meta-analysis framework. The method models each study’s activations under noise versus signal expectations and computes voxelwise LLRs to combine information asymmetrically across uneven datasets. Implemented in Python with NumPy and custom modules, it integrates watershed segmentation for cluster detection and simulation-based calibration. Validation compares results to ALE, testing robustness to injected noise and recovery from highly imbalanced datasets. Results/Anticipated Results: Studies are ongoing. We anticipate improved SNR and recovery robustness compared to ALE, as our approach avoids pre-thresholding before combining information across datasets. Simulation evidence suggests far greater computational efficiency (e.g., reducing 45-min–20-hour ALE runs to under two minutes locally). Conceptually, the framework parallels modern AI training strategies – modeling signal and noise distributions to calibrate evidence rather than relying on fixed significance thresholds. These innovations leverage information theory, watershed algorithms, and simulation calibration to yield interpretable, high-resolution evidence maps. Discussion/Significance of Impact: This framework advances coordinate-based meta-analysis by modeling evidence directly rather than inferring it through p-values. It improves reproducibility, interpretability, and cross-disorder comparability, supporting more robust and scalable neurocircuit mapping for researchers and clinicians.