Fig. 1: UMAP embedding of image properties and Histotyping features. | npj Precision Oncology

Fig. 1: UMAP embedding of image properties and Histotyping features.

From: Computer extracted gland features from H&E predicts prostate cancer recurrence comparably to a genomic companion diagnostic test: a large multi-site study

Fig. 1: UMAP embedding of image properties and Histotyping features.

Shown are UMAP projections based on (a) image properties and (b–g) the six features used in Histotyping on the entire validation set. Each dot represents a patient, and dots are colored according to: a, b the site the patient originated from, c Histotyping risk group, with kernel density, estimates shown for the Histotyping risk groups, d Gleason grade group, including kernel density estimates from (c), (e) Histotyping+ risk group, and (f) Decipher score, with markers indicating Histotyping+ risk group. g shows the patients who had BCR in less than one year or were BCR-free for more than 10 years, with outcomes indicated by marker shape. Regions of interest from select patients are shown from various locations in (g). Patients P1, P2, and P3 were Histotyping high-risk patients who had early BCR, while P4 and P5 were low-risk patients who were BCR free for more than 10 years.

Back to article page