Fast, Low-resource, and Accurate oRgan and Pan-cancer sEgmentation in Abdomen CT
Bibliographic record
Abstract
Abdomen organs are quite common cancer sites, such as colorectal cancer and pancreatic cancer, which are the 2nd and 3rd most common cause of cancer death [1,2]. Computed Tomography (CT) scanning yields important prognostic information for cancer patients and is a widely used technology for treatment monitoring. In both clinical trials and daily clinical practice, radiologists and clinicians measure the tumor and organ on CT scans based on manual two-dimensional measurements (e.g., Response Evaluation Criteria In Solid Tumors (RECIST) criteria) [3]. However, this manual assessment is inherently subjective with considerable inter- and intra-expert variability. Moreover, existing challenges mainly focus on one type of tumor (e.g., liver cancer, kidney cancer). There are still no general and publicly available models for universal abdominal organ and cancer segmentation at present. In this challenge, we aim to promote the development of universal organ and tumor segmentation in abdominal CT scans. This is an extension of FLARE2021 and FLARE2022 challenge. In FLARE2021, the challenge task is to segment four abdominal organs in a fully supervised setting. In FLARE2022, the challenge task is to segment 13 organs in a semi-supervised setting. In FLARE2023, we will add the lesion segmentation task. Different from existing tumor segmentation challenges [4,5], we focus on pan-cancer segmentation, which covers various abdominal cancer types. Specifically, the segmentation algorithm should segment 13 organs ( liver, spleen, pancreas, right kidney, left kidney, stomach, gallbladder, esophagus, aorta, inferior vena cava, right adrenal gland, left adrenal gland, and duodenum) and a single tumor class that includes all kinds of cancer types (such as liver cancer, kidney cancer, stomach cancer, pancreas cancer, colon cancer) in abdominal CT scans. Compared to the dataset in FLARE2021 (361 CT scans) and FLARE2022 (2300 CT scans), we further increase the dataset in FLARE2023 to 4500 CT scans (training/validation/testing: 4000, 100, 400), which is a multi-racial, multicenter, multi-disease, multi-phase, and multi-vendor dataset. To the best knowledge, this will be the largest and most diverse publicly available dataset for abdominal cancer analysis. The challenge will employ a partial-label learning setting where limited targets are annotated in each CT scan. This setting is in line with real-world settings because each medical department mainly focuses on one specific cancer. We aim to benchmark general abdominal cancer segmentation models that can handle multiple cancer types. Based on the results in FLARE2021 and FLARE2022, we found that segmentation models can achieve a good tradeoff between segmentation accuracy and efficiency. Thus, we will continue to evaluate both segmentation accuracy and efficiency. In particular, we will use Dice Similarity Coefficient (DSC), Normalized Surface Dice (NSD), and lesion-wise F1 score to evaluate segmentation accuracy, which is motivated by the metrics reloaded [6]. The segmentation efficiency is evaluated by running time and GPU memory consumption. In summary, the FLARE 2023 challenge has three main features: (1) Task: this is the first challenge for pan-cancer segmentation in abdominal CT scans (2) Dataset: we provide the largest abdomen CT dataset, including 4500 3D CT scans from 30+ medical centers. (3) Evaluation measures: we focus on both segmentation accuracy and segmentation efficiency [1] Nation Cancer Institute. "Cancer Stat Facts: Common Cancer Sites", 18 November 2022, https://seer.cancer.gov/statfacts/html/common.html. [2] Siegel, Rebecca, et al. "Cancer statistics, 2022. " CA: A Cancer Journal for Clinicians. 72 (2022): 7-33. [3] Eisenhauer, Elizabeth., et al. "New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1)." European journal of cancer 45.2 (2009): 228-247. [4] Bilic, Patrick, et al. The Liver Tumor Segmentation Benchmark (LiTS), Medical Image Analysis, (2022): 102680. [5] Heller, Nicholas, et al. "The state of the art in kidney and kidney tumor segmentation in contrast-enhanced CT imaging: Results of the KiTS19 challenge." Medical Image Analysis 67 (2021): 101821. [6] Maier-Hein, Lena, et al. "Metrics reloaded: Pitfalls and recommendations for image analysis validation." arXiv preprint arXiv:2206.01653 (2022).
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 | 0.001 | 0.004 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.003 | 0.002 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.004 | 0.002 |
| Insufficient payload (model declined to judge) | 0.003 | 0.004 |
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".