TB-MIL: deep learning-based identification of TMB status in bladder cancer from histopathological images
Bibliographic record
Abstract
Background and objective: Bladder cancer has an increasing prevalence, with non-muscle invasive cases frequently recurring and muscle-invasive cases exhibiting low survival rates. Recent advancements in cancer immunotherapy, particularly immune checkpoint inhibitors, offer promise but show variable patient responses, highlighting the need for predictive biomarkers like Tumor Mutation Burden (TMB). Although TMB correlates strongly with immunotherapy outcomes, traditional sequencing methods for TMB assessment are costly and time-consuming, necessitating more efficient alternatives. This study aimed to develop a deep learning model to accurately and cost-effectively predict TMB using bladder cancer pathology slides. Method: We retrospectively analyzed data from 354 bladder cancer patients in the TCGA database and 39 patients from our hospital, compiling clinical, pathological, and sequencing information. Our deep learning model, trained on segmented slide images, utilized various feature extractors and classifiers to optimize performance. Performance variances across subgroups based on T-stage and metastatic presence were statistically analyzed to assess the model’s stability across different stages. Model consistency was evaluated by comparing predictions from different slides of the same patient. Additionally, model visualization through attention mechanisms and GradCAM algorithm was conducted to analyze the reasons behind correct and incorrect TMB-H/L slide predictions, exploring potential pathological features related to TMB, and validating whether these features could guide human judgment of TMB levels. Results: The model achieved an AUC of 0.823 on an internal test set and 0.804 on an external test set, outperforming previous models. It demonstrated consistent accuracy across different T stages and metastasis statuses, with a high level of prediction consistency across multiple slides from the same patient. Visualization analyses revealed that the model focused on the intersection of tumor and non-tumor tissues, identifying distinct morphological features associated with high and low TMB levels. This information, when shared with clinicians, improved their TMB assessments. Conclusion: This study developed an efficient deep learning model, TB-MIL (TMB-BLCA-Multiple-Instance-Learning), for predicting TMB in bladder cancer from pathology slides, achieving high accuracy and consistency across multiple datasets and clinical subgroups. The model’s visualization provided valuable insights into TMB-associated morphological features, potentially enhancing clinical assessments.
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 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.002 | 0.001 |
| 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 | 0.001 |
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".