Perelman School of Medicine at the University of Pennsylvania

Section for Biomedical Image Analysis (SBIA)

participating with CBICA

HYDRA: Heterogeneity through Discriminative Analysis


Figure 1: Convex polytope classification used in HYDRA

 

HYDRA (Heterogeneity through Discriminative Analysis) is a joint multivariate classification and clustering technique that utilizes convex polytopes as classifiers (Figure 1). Its main aim is to address the question of whether there may be multiple subtypes of a disease which was once known to be unidirectional. This question is tackled by fitting a convex polytope classifier between the control group and the diseased group using either imaging or genetic features. The association of diseased group subjects to the faces of the polytope determine their memberships to different subtypes of disease patterns.

HYDRA was successfully applied for uncovering the underlying heterogeneity of Alzheimer’s disease using imaging patterns as well genetic signatures[1,2] (Figure 2). Also, HYDRA was used in the application of outlier detection and subtyping in a healthy aging population [3].

 

Figure 2: Three distinct subtypes of anatomical heterogeneity observed in Alzheimer's Disease

 

People

  • Christos Davatzikos
  • Erdem Varol
  • Aristeidis Sotiras

Publications

[1] Varol, Erdem, Aristeidis Sotiras, Christos Davatzikos, and Alzheimer's Disease Neuroimaging Initiative. "HYDRA: Revealing heterogeneity of imaging and genetic patterns through a multiple max-margin discriminative analysis framework." NeuroImage (2016).

[2] Varol, Erdem, Aristeidis Sotiras, and Christos Davatzikos. "Disentangling disease heterogeneity with max-margin multiple hyperplane classifier." In International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 702-709. Springer International Publishing, 2015.

[3]Varol, Erdem, Aristeidis Sotiras, and Christos Davatzikos. "Structured Outlier Detection in Neuroimaging Studies with Minimal Convex Polytopes." In International Conference on Medical Image Computing and Computer-Assisted Intervention, pp. 300-307. Springer International Publishing, 2016.