A Soft Labeling Approach to Develop Automated Algorithms that Incorporate Uncertainty in Pulmonary Opacification on Chest CT using COVID-19 Pneumonia
Bibliographic record
Abstract
Rationale and ObjectivesHard data labels for automated algorithm training are binary and cannot incorporate uncertainty between labels. We proposed and evaluated a soft labeling methodology to quantify opacification and percent well-aerated lung (%WAL) on chest CT, that considers uncertainty in segmenting pulmonary opacifications and reduces labeling burden.Materials and MethodsWe retrospectively sourced 760 COVID-19 chest CT scans from five international centers between January and June 2020. We created pixel-wise labels for >27,000 axial slices that classify three pulmonary opacification patterns: pure ground-glass, crazy-paving, consolidation. We also quantified %WAL as the total area of lung without opacifications. Inter-user hard label variability was quantified using Shannon entropy (range=0-1.39, low-high entropy/variability). We incorporated a soft labeling and modeling cycle following an initial model with hard labels and compared performance using point-wise accuracy and intersection-over-union of opacity labels with ground-truth, and correlation with ground-truth %WAL.ResultsHard labels annotated by 12 radiologists demonstrated large inter-user variability (3.37% of pixels achieved complete agreement). Our soft labeling approach increased point-wise accuracy from 60.0% to 84.3% (p=0.01) compared to hard labeling at predicting opacification type and area involvement. The soft label model accurately predicted %WAL (R=0.900) compared to the hard label model (R=0.856), but the improvement was not statistically significant (p=0.349).ConclusionOur soft labeling approach increased accuracy for automated quantification and classification of pulmonary opacification on chest CT. Although we developed the model on COVID-19, our intent is broad application for pulmonary opacification contexts and to provide a foundation for future development using soft labeling methods. Hard data labels for automated algorithm training are binary and cannot incorporate uncertainty between labels. We proposed and evaluated a soft labeling methodology to quantify opacification and percent well-aerated lung (%WAL) on chest CT, that considers uncertainty in segmenting pulmonary opacifications and reduces labeling burden. We retrospectively sourced 760 COVID-19 chest CT scans from five international centers between January and June 2020. We created pixel-wise labels for >27,000 axial slices that classify three pulmonary opacification patterns: pure ground-glass, crazy-paving, consolidation. We also quantified %WAL as the total area of lung without opacifications. Inter-user hard label variability was quantified using Shannon entropy (range=0-1.39, low-high entropy/variability). We incorporated a soft labeling and modeling cycle following an initial model with hard labels and compared performance using point-wise accuracy and intersection-over-union of opacity labels with ground-truth, and correlation with ground-truth %WAL. Hard labels annotated by 12 radiologists demonstrated large inter-user variability (3.37% of pixels achieved complete agreement). Our soft labeling approach increased point-wise accuracy from 60.0% to 84.3% (p=0.01) compared to hard labeling at predicting opacification type and area involvement. The soft label model accurately predicted %WAL (R=0.900) compared to the hard label model (R=0.856), but the improvement was not statistically significant (p=0.349). Our soft labeling approach increased accuracy for automated quantification and classification of pulmonary opacification on chest CT. Although we developed the model on COVID-19, our intent is broad application for pulmonary opacification contexts and to provide a foundation for future development using soft labeling methods.
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 | 0.002 | 0.002 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.002 | 0.004 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.002 |
| 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".