Unsupervised machine learning for risk stratification and identification of relevant subgroups of ascending aorta dimensions using cardiac CT and clinical data

Figures

• 

  Fig. 1. Cluster subgroups identified by unsupervised clustering. Plot showing 4 groups (Males A, Males B, Females A, Females B) of individuals identified by Random Forest and PAM (Partitioning Around Medoids) analysis.

• 

  Fig. 2. Comparison of variable importance in males (A) and Females (B) clusterization (from Random Forest model). We used 0.4 as the normalized weight threshold to select features used in the score definition. There is a difference on the x-axis between males and females because they have been analyzed separately and two different models (with two different feature importance) were built. AV, Aorta Variables; BV, Basic variables; RFV, Risk Factor Variables.

• 

  Fig. 3. Schematic representation of different ranges of aorta dimensions assigned to each cluster. A novel sample can be assigned to a cluster using their score compared to Th (threshold obtained from random forest feature importance analyses).

• 

  Figure S1. Comparison of Aorta dimension in males (A) and females (B) across the clusters. In all cases, there is a significant difference of values between cluster A and cluster B (p < 0.05)

• 

  Figure S2. Comparison of BMI (Body Mass Index), number of risk factor and age in males (A) and females (B) across the clusters. BMI was represented as Low (L) if < 25, High (H) if >=30 and Medium (M) if >= 24 and <30; Number of Risk Factors was represented as High (H) if >=3 and Low (L) if <3.