Skip to main content

Advertisement

Fig. 3 | Breast Cancer Research

Fig. 3

From: A whole slide image-based machine learning approach to predict ductal carcinoma in situ (DCIS) recurrence risk

Fig. 3

Full-slide feature selection for the development of recurrence classifier. a The change in model accuracy and high-risk group hazard ratio with the sequential addition of features. The reference hazard ratio and accuracies, based on the model with all features, are shown in red and blue horizontal dashed lines, respectively. The model which included all filtered features (Sig*: p < 0.05) is also shown for comparison. Bars on markers indicate 95% confidence intervals. b General feature descriptions, and the annotations from which they stem from, of the final 8-feature recurrence classification model. c Kaplan-Meier curves showing stratification of patient slides by the final recurrence classifier model. Data shown is based on the slides used for the training cohort, wherein the test sets for each selected cross-validated iteration were combined. Significance was measured using the log-rank test. d Univariate HR of the selected features, z-score transformed for illustrative purposes. All variables are significant, and blue horizontal lines depict 95% confidence intervals. The fact that none of the confidence intervals cross the HR = 1.0 reference line shows that these features are highly and unequivocally significant

Back to article page