Streamlining tuberculosis detection with foundation model-based weakly supervised transformer
Bibliographic record
Abstract
Tuberculosis (TB) remains a major global health challenge, particularly in low- and middle-income countries. Traditional microscopy-based diagnostics are labor-intensive and error-prone, while automated deep learning models often require detailed expert annotations and intensive preprocessing, limiting their scalability. To address these challenges, we propose a weakly supervised approach for detecting Mycobacterium tuberculosis (MTB) in microscopy images, leveraging UNI, a foundation model pretrained on millions of pathology images. Our method encodes microscopy images as sequences of patch-level embeddings using UNI and applies a Transformer encoder to classify each image using only image-level labels, without requiring detailed annotations. This framework minimizes preprocessing, reduces annotation costs, and enhances scalability. Our model was trained and tested on large, diverse datasets, achieving high PR-AUC scores (0.943-0.974), demonstrating strong performance and robustness. This success highlights the potential of our approach, which introduces two key innovations not previously explored for automated TB detection: leveraging cross-domain transfer learning by applying UNI for MTB detection and using a weakly supervised approach that relies only on image-level labels, significantly reducing the annotation burden compared to traditional fully supervised methods. Our results underscore the feasibility of foundation models in TB diagnostics and broader medical imaging applications. This scalable, weakly supervised approach demonstrates promising experimental results, highlighting its potential to significantly reduce annotation requirements and streamline TB detection workflows, particularly relevant to resource-limited settings.
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 distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 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 teacher head, 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".