Related Experiment Video
Updated: Aug 5, 2025
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
ADCoC: Adaptive Distribution Modeling Based Collaborative Clustering for Disentangling Disease Heterogeneity from
Hangfan Liu1, Michel J Grothe2, Tanweer Rashid3
1Neuroimage Analytics Laboratory (NAL) and Biggs Institute Neuroimaging Core, Glenn Biggs Institute for Neurodegenerative Disorders, University of Texas Health Science Center at San Antonio, San Antonio, Texas, USA; Center for Biomedical Image Computing and Analytics, University of Pennsylvania, Philadelphia, PA 19104, USA.
This study introduces a novel neuroimaging clustering method that simultaneously analyzes subjects and features, improving robustness against noisy data. The technique successfully identified distinct patient clusters in Parkinson's disease magnetic resonance imaging (MRI) data.
Area of Science:
- Neuroimaging analysis
- Machine learning in medicine
- Biostatistics
Background:
- Conventional clustering in neuroimaging often overlooks feature variability and data quality issues, leading to potential biases.
- Noisy neuroimaging data can significantly impact clustering accuracy and clinical interpretations.
- Existing methods frequently neglect the importance of feature grouping for optimizing subject clustering.
Purpose of the Study:
- To develop a robust neuroimaging clustering approach that addresses data noise and leverages feature heterogeneity.
- To improve the accuracy and reliability of subject clustering in neuroimaging studies.
- To identify distinct patient subgroups in Parkinson's disease using magnetic resonance imaging (MRI) data.
Main Methods:
- Simultaneous clustering of subjects and features using nonnegative matrix tri-factorization.
- Introduction of adaptive regularization based on coefficient distribution modeling to suppress noise.
- Development of a noise-robust method by modeling non-negative coefficient distributions tailored to the data.
Main Results:
- The proposed method demonstrated superior clustering performance on synthetic data compared to standard and recent techniques.
- Application to Parkinson's disease MRI data revealed two stable, reproducible patient clusters.
- These identified clusters were characterized by distinct patterns of cortical/medial temporal atrophy and associated cognitive differences.
Conclusions:
- The novel nonnegative matrix tri-factorization approach with adaptive regularization enhances clustering robustness and accuracy in noisy neuroimaging data.
- The method effectively identifies clinically relevant subgroups in neurodegenerative diseases like Parkinson's disease.
- This technique offers a promising tool for advancing precision medicine in neurology through improved data analysis.
