Skip to main content
Fig. 4 | Breast Cancer Research

Fig. 4

From: Deep learning-based risk stratification of preoperative breast biopsies using digital whole slide images

Fig. 4

DeepGrade prediction results obtained on biopsy specimens compared to the clinical Nottingham Histological Grade (NHG) assigned by the pathologist on the resected specimen. A. Example of a whole slide image with prediction results. Red is more likely to be predicted as high risk, or in other words the predicted probability for each tile to be classified as high-risk. The patient is classified DeepGrade high as the upper percentile of the mean values across all tiles was over the obtained threshold of 0.83. B. Receiver Operating Curve (ROC) of the patient-level DeepGrade model prediction versus the resected tumour grades NHG1 and NHG3 assigned by a pathologist. C. Sankey plot of the proportion of patients predicted with DeepGrade-high and -low versus the resected tumour grade NHG1 and NHG3. D. Sankey plot with results of all biopsy specimen comparing the obtained DeepGrade risk group with both the biopsy NHG and the resected tumour NHG. Diagnoses were in agreement when patients were low-risk biopsy DeepGrade and resected tumour NHG1 or high-risk biopsy DeepGrade and resected tumour NHG3. Patients with resected tumour NHG2 were stratified as either low-risk or high-risk

Back to article page