Abstract PO-031: Evaluating clinical utility of organs-at-risk segmentation in head & neck cancer by simple open-source 3D CNNs
Bibliographic record
Abstract
Abstract Open source code helps the scientific community converge to breakthroughs at an accelerated rate. Only one of 14 previous deep learning based OAR segmentation studies met most open-source standards. Using this study as a benchmark, we will assess the segmentation quality of each network discussed and provide the community with pre-trained weights of the top-performing models. 11 open-source 3D segmentation models originally engineered for medical image segmentation in both the 2D and 3D domains were trained to automatically segment 19 OaR classes. For this study, a large internally curated dataset from the University Health Network (UHN) of 582 patient scans was used. All models were tested on a hold out set from the 582 cohort of 59 patients. Models were also tested on 98 external patient scans from publicly available sources. 10 test set contours from the winning network were assessed by an expert radiation oncologist with 10+ years of experience to identify the observer of the contour set (whether AI or human). Preliminary results show that only 13 out of 20 scans were identified correctly. Results show that simple 3D architectures consistently outcompete more complex networks by producing more accurate, clinically acceptable contours. A more thorough clinical acceptability test is underway to establish a protocol for integrating deep learning based auto contouring models into radiation therapy planning workflows. Citation Format: Joseph Marsilla, Benjamin Haibe-Kains. Evaluating clinical utility of organs-at-risk segmentation in head & neck cancer by simple open-source 3D CNNs [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-031.
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 | 0.011 |
| 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 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.002 | 0.001 |
| Insufficient payload (model declined to judge) | 0.004 | 0.002 |
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".