Pursuit of biomarkers of brain diseases: Beyond cohort comparisons

In a new arXiv paper titled 'Pursuit of biomarkers of brain diseases: Beyond cohort comparisons,' researchers Pascal Helson and Arvind Kumar deliver a critical analysis of why AI and big data have failed to translate into clinically useful brain disease biomarkers. They argue the field is stuck on a flawed paradigm: comparing groups of patients to healthy controls using single data types like fMRI or EEG. The authors introduce a 'Brain Swap' thought experiment to illustrate that even with infinite data and advanced algorithms, this approach cannot overcome the fundamental 'degeneracy' of brain features—where different neural configurations can produce the same function or behavior.

Helson and Kumar propose a complete methodological shift. Instead of seeking a single biomarker from one data source, they advocate for a multimodal, longitudinal framework. This means simultaneously tracking brain activity, neurotransmitter levels, neuromodulators, and structural imaging over time in individuals. The goal is to first use this rich, personalized data to guide the grouping of patients into more biologically meaningful categories, and then define multidimensional biomarkers from these patterns. This approach moves beyond simple correlation to capture the dynamic, systems-level dysfunction underlying conditions like Alzheimer's or Parkinson's.