Self-supervised learning and uncertainty estimation for surgical margin detection
Bibliographic record
Abstract
Up to 35% of breast-conserving surgeries fail to resect all the tumors completely. Ideally, machine learning methods using the iKnife data, which uses Rapid Evaporative Ionization Mass Spectrometry (REIMS), can be utilized to predict tissue type in real-time during surgery, resulting in better tumor resections. As REIMS data is heterogeneous and weakly labeled, and datasets are often small, model performance and reliability can be adversely affected. Self-supervised training and uncertainty estimation of the prediction can be used to mitigate these challenges by learning the signatures of input data without their label as well as including predictive confidence in output reporting. We first design an autoencoder model using a reconstruction pretext task as a self-supervised pretraining step without considering tissue type. Next, we construct our uncertainty-aware classifier using the encoder part of the model with Masksembles layers to estimate the uncertainty associated with its predictions. The pretext task was trained on 190 burns collected from 34 patients from Basal Cell Carcinoma iKnife data. The model was further trained on breast cancer data comprising of 200 burns collected from 15 patients. Our proposed model shows improvement in sensitivity and uncertainty metrics of 10% and 15.7% over the baseline, respectively. The proposed strategies lead to improvements in uncertainty calibration and overall performance, toward reducing the likelihood of incomplete resection, supporting removal of minimal non-neoplastic tissue, and improved model reliability during surgery. Future work will focus on further testing the model on intraoperative data and additional exvivo data following collection of more breast samples.
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 | 0.002 | 0.006 |
| 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 | 0.001 |
| 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".