NiChart ML Models
NiChart offers an extensible library of pre-trained machine learning (ML) models that can convert high-dimensional imaging data into low-dimensional imaging signatures. These representations effectively capture and quantify brain changes associated with specific diseases or neurodegenerative conditions.
The collection of NiChart imaging signatures contributes to the neuroimaging chart dimensional system. NiChart's pre-trained ML models are readily available, thereby eliminating the need for extensive training or expertise in machine learning. Additionally, the extensibility of the NiChart library will allow researchers to add their own specialized models, after harmonizing their data with NiChart.
The models are trained on carefully selected subsets of the reference dataset, tailored to each task and target disease/condition.
SPARE or Spatial Patterns of Abnormality for Recognition of Disease models are predictive supervised learning methods that have been extensively validated. SPARE models train on imaging features extracted from single or multi-modal MRI scans. The models use these features to learn how to identify patterns in the brain that are associated with different diseases. Initial models are provided for SPARE-BA (brain age) and SPARE-AD (Alzheimer's disease). Additional models for SPARE-CVD (cardio-vascular disease risk), SPARE-DM (Type2 diabetes), SPARE-SCZ (schizophrenia) and SPARE-CD (chronic depression) as well as models derived from weakly supervised methods and which identify subtypes of these diseases, will be added in future releases.
Figure 1. Grey matter and white matter group differences between individuals with low vs high SPARE-AD values (from 1).
Our research team has developed ML tools to uncover imaging patterns of disease heterogeneity from MRI data. These tools help us identify distinct disease subtypes that shed light on the underlying neuroanatomical differences associated with various pathologies. Our previous work has identified four distinct disease subtypes for Alzheimer's disease and two subtypes for schizophrenia. The pre-trained models provided in NiChart will enable users to obtain more nuanced measures beyond the traditional disease scores.
Figure 2. Alzheimer's disease subtypes identified by the SMILE-GAN method (from 2).