Multicenter Imaging Studies: Automated Approach to Evaluating Data Variability and the Role of Outliers
Bibliographic record
Abstract
Magnetic resonance (MR) as well as other imaging modalities have been used in a large number of clinical and research studies for the analysis and quantification of important structures and the detection of abnormalities. In this context, machine learning is playing an increasingly important role in the development of automated tools for aiding in image quantification, patient diagnosis and follow-up. Normally, these techniques require large, heterogeneous datasets to provide accurate and generalizable results. Large, multi-center studies, for example, can provide such data. Images acquired at different centers, however, can present varying characteristics due to differences in acquisition parameters, site procedures and scanners configuration. While variability in the dataset is required to develop robust, generalizable studies (i.e., independent of the acquisition parameters or center), like all studies there is also a need to ensure overall data quality by prospectively identifying and removing poor-quality data samples that should not be included, e.g., outliers. We wish to keep image samples that are representative of the underlying population (so called inliers), yet removing those samples that are not. We propose a framework to analyze data variability and identify samples that should be removed in order to have more representative, reliable and robust datasets. Our example case study is based on a public dataset containing T1-weighted volumetric head images data acquired at six different centers, using three different scanner vendors and at two commonly used magnetic fields strengths. We propose an algorithm for assessing data robustness and finding the optimal data for study occlusion (i.e., the data size that presents with lowest variability while maintaining generalizability (i.e., using samples from all sites)).
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 | 0.066 | 0.121 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.003 | 0.003 |
| Bibliometrics | 0.011 | 0.008 |
| Science and technology studies | 0.002 | 0.002 |
| Scholarly communication | 0.005 | 0.003 |
| Open science | 0.004 | 0.005 |
| Research integrity | 0.003 | 0.002 |
| Insufficient payload (model declined to judge) | 0.001 | 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 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".