Skip to main content

Advertisement

Fig. 4 | Breast Cancer Research

Fig. 4

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

Fig. 4

Interpretation and prognostic relevance of the most prognostic feature in our eight-feature DCIS recurrence risk prediction model. a An example “cancer” region with a cribriform architecture in an H&E-stained slide (prior to deconvolution). b The region shown in a after hematoxylin deconvolution. c Intense hematoxylin staining (relative to the image tile section) is represented by a gray-level intensity of 1, while no staining is depicted by a gray-level value of 255. The adaptive Otsu thresholds by progressively using a higher threshold. Therefore, if the cancer region has lumens, it would yield a higher average intensity (more white pixels) as compared to a solid pattern (no white pixels). Using an optimized threshold of 208, it is observed that full slides whose cancer regions have an average feature #1 above that cutoff recur significantly less than patients below that threshold (d)

Back to article page