Abstract PO-010: Feature pyramid network for revealing tumour infiltrating lymphocyte presence and distribution in a whole slide image
Bibliographic record
Abstract
Abstract Ductal Carcinoma in Situ (DCIS) is a subtype of non-invasive breast cancer contained in the ducts of the breast. DCIS is not a life-threatening condition however a small number of cases will progress to or reoccur as invasive breast cancer. Clinical workflows lack a robust biomarker to determine which patients will reoccur. Recently pathologists have turned towards investigating the tumour microenvironment to predict cancer recurrence. Machine learning models present an opportunity to analyze these features on large amounts of data with minimal requirement of pathologist time. One feature of recent interest are Tumour infiltrating lymphocytes or TILs, lymphocytes within a small region around a cancerous duct. Through a method of bootstrapping a U-Net on top of a fully convolutional network we have designed a network that is able to segment lymphocyte and malignant cells in H&E Images of Ductal Carcinoma In Situ. The original network was trained on a private set of cellularity data in breast cancer, manually annotated by pathologists to mark the centre and type of each cell in a patch. Probabilistic output from this model was used to generate segmentation maps on a much larger dataset of DCIS images from the Ontario DCIS Cohort. These segmentation maps were used as training data for a feature pyramid network. The results of this bootstrapping method was a significant increase to speed of the model in addition to much larger flexibility in the input size and shape for the model, which came at a small cost to overall model accuracy. In addition to the original segmentation target, the use of this model to generate lower resolution WSI heat-maps has been investigated for this work. This model can generate low resolution heat-maps for a 20X magnification whole slide image showing the presence of malignant cells in addition to the location and distribution of TILs around ducts. This can provide a visual overview of differences between slides and we will be examining the potential for this output to serve as information for analytical methods to process to predict recurrence free survival. Citation Format: Jonathan Mazurski, Sharon Nofech-Mozes, Dina Bassiouny, Anne L. Martel. Feature pyramid network for revealing tumour infiltrating lymphocyte presence and distribution in a whole slide image [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-010.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.001 | 0.000 |
| Insufficient payload (model declined to judge) | 0.004 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".