Deep Learning-Based Gleason Grading of Prostate Cancer From Histopathology Images—Role of Multiscale Decision Aggregation and Data Augmentation
Bibliographic record
Abstract
Visual inspection of histopathology images of stained biopsy tissue by expert pathologists is the standard method for grading of prostate cancer (PCa). However, this process is time-consuming and subject to high inter-observer variability. Machine learning-based methods have the potential to improve efficient throughput of large volumes of slides while decreasing variability, but they are not easy to develop because they require substantial amounts of labeled training data. In this paper, we propose a deep learning-based classification technique and data augmentation methods for accurate grading of PCa in histopathology images in the presence of limited data. Our method combines the predictions of three separate convolutional neural networks (CNNs) that work with different patch sizes. This enables our method to take advantage of the greater amount of contextual information in larger patches as well as greater quantity of smaller patches in the labeled training data. The predictions produced by the three CNNs are combined using a logistic regression model, which is trained separately after the CNN training. To effectively train our models, we propose new data augmentation methods and empirically study their effects on the classification accuracy. The proposed method achieves an accuracy of [Formula: see text] in classifying cancerous patches versus benign patches and an accuracy of [Formula: see text] in classifying low-grade (i.e., Gleason grade 3) from high-grade (i.e., Gleason grades 4 and 5) patches. The agreement level of our automatic grading method with expert pathologists is within the range of agreement between pathologists. Our experiments indicate that data augmentation is necessary for achieving expert-level performance with deep learning-based methods. A combination of image-space augmentation and feature-space augmentation leads to the best results. Our study shows that well-designed and properly trained deep learning models can achieve PCa Gleason grading accuracy that is comparable to an expert pathologist.
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 | 0.001 | 0.004 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
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".