Self-Supervised Learning for Rare Disease Diagnosis Using Limited and Imbalanced Medical Imaging Data
Bibliographic record
Abstract
This research depends on a self-supervised learning framework as its principal methodology to enhance rare disease medical imaging diagnoses through minimal data sources that exhibit significant imbalance challenges. The proposed framework achieves accurate classification of underrepresented medical anomalies through its combination of SSL representational power together with fine-grained class rebalancing and cross-modal pretext tasks. The diagnostic system maintains interpretability through visualization methods and mapping activations at each class to enable medical analysis despite limited annotated medical data. The proposed research adds three essential components to medical imaging analysis that comprise (1) a contrastive learning encoder which extracts knowledge from unlabeled patient data to represent rare diseases, (2) a distribution-aware balancing method that corrects class bias and (3) a clinical validation process that improves diagnostic accuracy by 19% over standard CNN and transformer-based models while delivering$\mathbf{3 5 \%}$better F1-scores in rare class diagnosis. The SSL-based diagnostic pipeline reduces false-negative results through its mechanism which lowers detection errors by 24% thereby improving trust and operational efficiency. The research achieves improved medical AI system performance by integrating scalable self-supervised learning with interpretable diagnosis methods for use in low-resource clinical situations where risks are high. The research uses SSL approaches to process medical imaging data with imbalance while achieving optimal disease diagnosis through small expert-inputs.
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 | 0.003 | 0.008 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.002 | 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".