34 — Yet Another ADNI Machine Learning Paper? Paving The Way Towards Fully-reproducible Research on Classification of Alzheimer’s Disease

This paper calls attention to the large number of recent irreproducible Alzheimer’s (AD) ML-diagnosis papers, and laments the lack of interoperability between these datasets, classifiers, and conclusions. While many research studies use the public ADNI data, the authors point out that ADNI is a heterogenous mix of patients, preprocessing pipelines, and metrics. To remedy this, the authors provide a way to convert ADNI brains into the BIDS format, another standard in brain-imaging.

The authors then provide a plug-and-play architecture for a BYO-classifier and feature extractor. This means that anyone can design their own algorithms and run them on a (more) homogenous BIDS dataset, enabling reproducible AD science.

In particular, I like how accessible the code is, both in terms of availability and in code-quality.