Abstract WP406: Training and Validation of Deepmedic Machine Learning Tool for Automated Hematoma Segmentation and Volume Analysis on CT Using Multicenter Data
Bibliographic record
Abstract
Introduction: We sought to train and validate an automated machine learning algorithm for ICH segmentation and volume calculation using multicenter data. Methods: An open-source 3D deep machine learning algorithm “DeepMedic” was trained using manually segmented ICH from 208 CT scans (129 patients) from the multicenter PREDICT study. The algorithm was then validated with 125 manually segmented CT scans (48 patients) from the SPOTLIGHT study. Manual segmentation was performed with Quantomo semi-automated software. ABC/2 was measured for all studies by two neuroradiologists. Accuracy of DeepMedic segmentation was assessed using the Dice similarity coefficient. Analysis was stratified by presence of IVH. Intraclass correlation (ICC) with 95% confidence intervals (CI) assessed agreement between manual vs. DeepMedic segmentation volume; and manual segmentation and ABC/2 volume. Bland-Altman charts were analyzed for ABC/2 and DeepMedic vs. manual segmentation volumes. Results: DeepMedic demonstrated high segmentation accuracy in the training cohort (median Dice 0.96; IQR 0.95 - 0.97) and in the validation cohort (median Dice 0.91; IQR 0.86 - 0.94). Dice coefficients were not significantly different between patients with IVH in the training cohort; however was significantly worse in the validation cohort in patients with IVH (Wilcoxon p<0.001). Agreement was significantly better between DeepMedic and manual segmentation (PREDICT: ICC 0.99 [95%CI 0.99 -1.00]; SPOTLIGHT: ICC 0.98 [95%CI 0.97 - 0.99]) than between ABC/2 and manual segmentation (PREDICT: ICC 0.92 [95%CI 0.89 - 0.95]; SPOTLIGHT: ICC 0.95 [95%CI 0.93-0.97]). Improved accuracy of DeepMedic was demonstrated in Bland-Altman charts (Fig 1). Conclusion: ICH machine learning segmentation with DeepMedic is feasible and accurate; and demonstrates greater agreement with manual segmentation compared to ABC/2 volumes. Accuracy of the machine learning algorithm however is limited in patients with IVH.
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 | 0.008 | 0.016 |
| 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 | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 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".