Real-time AI-powered EUS imaging analysis software for detection and segmentation of cystic versus solid pancreatic lesions
Bibliographic record
Abstract
Aims Detection and differentiation of cystic versus solid pancreatic lesions prior to performance of EUS-guided fine needle aspiration biopsy (EUS-FNA/B) can be automated by use of artificial intelligence (AI) techniques based on convolutional neural networks (CNN). We aimed to test the technical feasibility of a novel real-time AI-powered EUS imaging analysis software for detection and segmentation of the pancreas, cystic pancreatic lesions and/or solid pancreatic masses (PANC-AI). Methods 202 consecutive patients undergoing EUS examination of pancreas were included in the study. Two expert EUS investigators labelled the EUS grey-scale movies with respect to identification of pancreas (uncinate, head, neck, body and tail), cystic pancreatic lesions and solid pancreatic masses. The gold standard for diagnosis of cystic and/or solid pancreatic lesions was achieved by EUS-FNA/B with rapid on-site assessment followed by histology. Individual images (frames) were uploaded into VGG Image Annotator (https://www.robots.ox.ac.uk/~vgg/software/via/) and cross-labelled by two independent physician annotators. Training was performed using a faster but less complex CNN1 and a slower but more complex CNN2, based on the first 150 patients (29586 frames). Subsequent testing was performed on 52 different patients (3127 frames). The movies tested with the CNNs were further categorized by non-clinician operators into the same diagnostic categories. Results Based on acquired preliminary data, we compared the percentage (ratio) of EUS procedures that yield clinically relevant information for the detection of cystic and/or solid pancreatic lesions in a typical patient population submitted for pancreas EUS. The overall accuracy for case-based analysis was 100% (CNN1) & 91% (CNN2) for cystic pancreatic lesions, and 100% (CNN1) & 92% (CNN2) for solid pancreatic masses, respectively. The overall accuracy, precision and recall for frame-based analysis were automatically calculated and reported, being 77%, 91%, 57% (CNN1) & 80%, 72%, 63% (CNN2) for cystic pancreatic lesions, as well as 77%, 83%, 62% (CNN1) & 75%, 75%, 72% (CNN2) for solid pancreatic masses, respectively. Conclusions In conclusion, the PANC-AI software reliably detected pancreas and segmented all cystic pancreatic lesions and/or solid pancreatic masses before the performance of EUS-FNA/B. With additional refinements, this technology may have the potential to improve learning curve and operating characteristics of EUS for pancreas examinations. Publication History Article published online: 15 April 2024 © 2024. European Society of Gastrointestinal Endoscopy. All rights reserved. Georg Thieme Verlag KG Rüdigerstraße 14, 70469 Stuttgart, Germany
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 | 0.010 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.000 |
| Insufficient payload (model declined to judge) | 0.004 | 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".