Bayesian lesion-deficit inference with Bayes factor mapping: Key advantages, limitations, and a toolbox.

Area of Science:

• Neuroscience
• Neurology
• Biostatistics

Background:

• Frequentist lesion-symptom mapping (LSM) is standard for brain lesion analysis but faces limitations like the multiple comparison problem, association problem, and low statistical power.
• Frequentist methods lack insight into evidence for the null hypothesis (absence of effects).
• Bayesian lesion deficit inference (BLDI) offers a potential improvement by providing evidence for the null hypothesis and avoiding accumulated errors.

Purpose of the Study:

• To implement and evaluate Bayesian lesion deficit inference (BLDI) using Bayes factor mapping.
• To compare the performance of BLDI against traditional frequentist lesion-symptom mapping.
• To assess BLDI's utility in identifying lesion-deficit associations and evidence for the null hypothesis.

Main Methods:

• Implemented BLDI using Bayes factor mapping with Bayesian t-tests and general linear models.
• Compared BLDI with frequentist LSM using permutation-based family-wise error correction.
• Evaluated methods in an in-silico study (300 stroke patients) and a clinical study (137 stroke patients) for voxel-wise and disconnection-wise analyses.

Main Results:

• BLDI identified areas with evidence for the null hypothesis and was more liberal in detecting lesion-deficit associations.
• BLDI outperformed frequentist methods with small lesions and low statistical power, offering greater transparency.
• BLDI showed increased false positives (association problem) in high-power analyses; adaptive lesion size control mitigated this.

Conclusions:

• BLDI is a valuable addition to lesion-deficit inference methods, especially for small lesions, low power, and identifying absent associations.
• BLDI is not a universal replacement for frequentist approaches but offers unique advantages.
• An R toolkit for BLDI analysis was published to enhance accessibility.