Model-based clustering of multiple images incorporating covariates

Area of Science:

• Biostatistics
• Medical Imaging
• Public Health

Background:

• Anemia screening is crucial, especially in low-resource settings.
• Current methods can be invasive or costly.
• Smartphone-based diagnostics offer a promising alternative.

Purpose of the Study:

• To develop a novel functional data clustering method for image analysis.
• To identify anemia risk subgroups using smartphone-derived fingernail images.
• To adjust for covariate effects on cluster membership.

Main Methods:

• Representing images as two-dimensional functional data.
• Formulating a functional latent class mixed model.
• Applying the method to fingernail color intensity matrices, adjusting for metadata.

Main Results:

• Identified three distinct subgroups with varying anemia risk.
• Cluster 1: 0% anemic, Cluster 2: 79% anemic, Cluster 3: 86% anemic.
• Demonstrated the method's utility in identifying high-risk populations.

Conclusions:

• The proposed functional data clustering method is effective for image analysis.
• Smartphone-based fingernail image analysis is a viable tool for non-invasive anemia screening.
• This approach can facilitate targeted public health interventions for anemia.