Philippe Weitz

Introduction: Nottingham histological grade (NHG) is a well established prognostic factor in breast cancer histopathology. However, manual NHG assessment of biopsies is challenging and has a large inter-assessor variability with a large proportion being classified as NHG2 (intermediate grade). Here, we evaluate whether DeepGrade, a previously developed model for the risk stratification of resected tumour specimens, could be applied to risk-stratify biopsy specimens. Methods: A total of 11,943,905 tiles from 1171 whole slide images (WSIs) of preoperative biopsies from 897 patients diagnosed with breast cancer in Stockholm, Sweden, were included in this retrospective observational study. DeepGrade, a deep convolutional neural network model, was applied for classification of low and high risk tumours and evaluated against clinically assigned grades 1 and 3 using area under the operating curve (AUC). The prognostic value of the DeepGrade model in the biopsy setting was evaluated using time-to-event analysis. Results: The DeepGrade model classified resected tumour cases with grades NHG1 and NHG3 using only biopsy specimens with an AUC of 0.903 (95% CI: 0.88;0.93). The model could also classify the biopsy NHG (1 and 3) assessed on the biopsy of 186 patients with an AUC of 0.959 (95% CI: 0.93; 0.99). Furthermore, out of the 434 NHG2 tumours, 255 (59%) were classified as DeepGrade2-low, and 179 (41%) were classified as DeepGrade2-high. Using a multivariable Cox proportional hazards model the hazard ratio between low- and high-risk groups was estimated as 2.01 (p-value = 0.036). Conclusions: DeepGrade could predict the resected tumour grades NHG1 and NHG3 using only the biopsy specimen and sub-classify grade 2 tumours into low and high risks. The results demonstrate that the DeepGrade model can provide decision support for biopsy grading, and potentially provide decision support in the clinical setting to identifying high-risk tumours based on preoperative breast biopsies, thus improving information available for clinical treatment decisions.

The analysis of FFPE tissue sections stained with haematoxylin and eosin (H&E) or immunohistochemistry (IHC) is essential for the pathologic assessment of surgically resected breast cancer specimens. IHC staining has been broadly adopted into diagnostic guidelines and routine workflows to assess the status of several established biomarkers, including ER, PGR, HER2 and KI67. Biomarker assessment can also be facilitated by computational pathology image analysis methods, which have made numerous substantial advances recently, often based on publicly available whole slide image (WSI) data sets. However, the field is still considerably limited by the sparsity of public data sets. In particular, there are no large, high quality publicly available data sets with WSIs of matching IHC and H&E-stained tissue sections from the same tumour. Here, we publish the currently largest publicly available data set of WSIs of tissue sections from surgical resection specimens from female primary breast cancer patients with matched WSIs of corresponding H&E and IHC-stained tissue, consisting of 4,212 WSIs from 1,153 patients.

The alignment of tissue between histopathological whole-slide-images (WSI) is crucial for research and clinical applications. Advances in computing, deep learning, and availability of large WSI datasets have revolutionised WSI analysis. Therefore, the current state-of-the-art in WSI registration is unclear. To address this, we conducted the ACROBAT challenge, based on the largest WSI registration dataset to date, including 4,212 WSIs from 1,152 breast cancer patients. The challenge objective was to align WSIs of tissue that was stained with routine diagnostic immunohistochemistry to its H&E-stained counterpart. We compare the performance of eight WSI registration algorithms, including an investigation of the impact of different WSI properties and clinical covariates. We find that conceptually distinct WSI registration methods can lead to highly accurate registration performances and identify covariates that impact performances across methods. These results establish the current state-of-the-art in WSI registration and guide researchers in selecting and developing methods.

Online-only supplementary tables: Table S1. Clinicopathological characteristics of patients per cohort. Table S2. ST gene panel. Table S3. The list of features used in the neural network based classification. Table S4. Results of pre-study hyperparameter tunning on 5 genes. Table S5. Prediction performance for the 1011 genes on validation and test sets. Table S6. Prediction performance for transcripts from established biomarker panels. Table S7. Correlation of gene expression with ER status (internal test set). Table S8. Correlation of gene expression with ER status (external test set). Table S9. Enriched pathways for genes cannot be predicted (Reactome gene sets). Table S10. Enriched pathways for genes cannot be predicted (Hallmark gene sets). Table S11. List of proliferation genes.

Online-only supplementary tables: Table S1. Clinicopathological characteristics of patients per cohort. Table S2. ST gene panel. Table S3. The list of features used in the neural network based classification. Table S4. Results of pre-study hyperparameter tunning on 5 genes. Table S5. Prediction performance for the 1011 genes on validation and test sets. Table S6. Prediction performance for transcripts from established biomarker panels. Table S7. Correlation of gene expression with ER status (internal test set). Table S8. Correlation of gene expression with ER status (external test set). Table S9. Enriched pathways for genes cannot be predicted (Reactome gene sets). Table S10. Enriched pathways for genes cannot be predicted (Hallmark gene sets). Table S11. List of proliferation genes.

ST analysis results for Tumour(PART1)

Plot of delta area showing the relative change in area under the cumulative distribution function (CDF) curve from k-1 to k.

Supplementary comparison of model performance

Heat maps showing clusters of patients based on 44 genes (FDR adjusted p-value < 0.05) from the PAM50 gene panel.