MétaCan
Menu
Back to cohort

Abstract PR-07: Automated segmentation pipeline for radiological imaging using UniverSeg and similarity-guided support set retrieval

2025· article· en· W4412163707 on OpenAlexaboutno aff
Niket Patel, Vishwa S. Parekh

Bibliographic record

VenueClinical Cancer Research · 2025
Typearticle
Languageen
FieldMedicine
TopicRadiomics and Machine Learning in Medical Imaging
Canadian institutionsnot available
Fundersnot available
KeywordsPipeline (software)Radiological weaponSegmentationSimilarity (geometry)Artificial intelligenceMedicineComputer scienceSet (abstract data type)Pattern recognition (psychology)Computer visionRadiologyImage (mathematics)

Abstract

fetched live from OpenAlex

Abstract Automated segmentation is essential in medical imaging, aiding the identification of organs and pathologies for diagnosis, treatment planning, and disease monitoring. However, traditional models rely on large, manually-labeled datasets, task-specific retraining, and complex setup processes, limiting scalability in clinical practice. These challenges are particularly significant in oncology, where timely, accurate segmentation of tumors and affected organs is crucial. This project introduces a scalable, AI-driven pipeline centered around UniverSeg, a pretrained model that performs multi-structure segmentation with minimal user input. UniverSeg leverages a small support set of annotated images, eliminating the need for retraining and enabling adaptability across diverse imaging tasks.A key innovation of this pipeline is the automatic selection of an optimal support set. Searching algorithms, including hash-based methods, structural similarity index (SSIM), learned perceptual image patch similarity (LPIPS), and dense vector comparison, retrieve annotated images from a pre-built database (e.g., TotalSegmentator with 1,228 CT scans and 117 segmented structures) that closely match the input image provided by the radiologist. The retrieved support set then guides UniverSeg in accurately segmenting multiple structures within the scan.We evaluated the performance of various search methods and support set sizes (5, 10, and 15 patients). LPIPS delivered the highest segmentation accuracy for complex anatomical regions, while hash-based and dense vector methods demonstrated efficient image retrieval. Larger support sets further improved segmentation performance, particularly for intricate regions such as lungs and abdominal organs. The pipeline achieved consistent results across TotalSegmentator and BTCV datasets. This solution has significant implications for oncology workflows. Accurate segmentation of tumors and related structures is essential for assessing tumor burden, quantifying volumes, planning radiation therapy, and tracking treatment response. For example, a radiologist can upload a CT slice containing a tumor, and the pipeline will automatically segment the tumor and surrounding structures, providing critical data for precision treatment. The system also reduces variability across institutions and enhances workflow efficiency by eliminating extensive manual annotations and specialized AI requirements. By integrating UniverSeg with automated support set retrieval, this pipeline bridges the gap between AI research and clinical oncology. Its ability to streamline and scale segmentation tasks improves imaging analysis, supporting better decision-making in cancer diagnosis, prognosis, and treatment planning. Citation Format: Niket Patel, Vishwa Parekh. Automated segmentation pipeline for radiological imaging using UniverSeg and similarity-guided support set retrieval [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Artificial Intelligence and Machine Learning; 2025 Jul 10-12; Montreal, QC, Canada. Philadelphia (PA): AACR; Clin Cancer Res 2025;31(13_Suppl):Abstract nr PR-07.

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 imitation

Not 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.

metaresearch head score (Codex)0.006
metaresearch head score (Gemma)0.005
Version: codex-gemma-dda1882f352aValidation status: machine_predicted_unvalidated
Candidate categoriesnone
Consensus categoriesnone
DomainCandidate signal: none · Consensus signal: none
Study designCandidate signal: Simulation or modeling · Consensus signal: none
GenreCandidate signal: Empirical · Consensus signal: Empirical
Teacher disagreement score0.845
Threshold uncertainty score0.643

Codex and Gemma teacher scores by category

CategoryCodexGemma
Metaresearch0.0060.005
Meta-epidemiology (narrow)0.0000.000
Meta-epidemiology (broad)0.0000.000
Bibliometrics0.0000.000
Science and technology studies0.0000.001
Scholarly communication0.0000.000
Open science0.0000.000
Research integrity0.0000.001
Insufficient payload (model declined to judge)0.0000.000

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.

Opus teacher head0.233
GPT teacher head0.579
Teacher spread0.345 · how far apart the two teachers sit on this one work
Validation statusscore_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it

Classification

machine, unvalidated

Machine predicted; a candidate call from one teacher head, not a consensus.

The models applied no category: nothing in the taxonomy fit this work.
Study designSimulation or modeling
Domainnot available
GenreEmpirical

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".

Quick stats

Citations2
Published2025
Admission routes1
Has abstractyes

Explore more

Same venueClinical Cancer ResearchSame topicRadiomics and Machine Learning in Medical ImagingFrench-language works237,207