Semi-supervised active transfer learning for fetal ECG arrhythmia detection
Bibliographic record
Abstract
Deep learning has demonstrated excellent results for ECG anomaly detection, wherein most approaches used supervised learning. The requirement of thousands of manually annotated samples is a concern for state-of-the-art anomaly detection systems, especially for fetal ECG (FECG), and currently, there is not a publicly available FECG dataset annotated for each FECG beat. In this paper, we offer a modified active learning technique based on transfer learning, calibration probability, and autoencoder-based sampling to reduce number of samples requires to annotate. In this regard, we used 25,000 s of recording from 47 patients from the MIT-BIH Arrhythmia Database to train a deep learning model to detect anomalies in non-fetus subjects. Then we used the unlabeled Non-Invasive Fetal ECG Arrhythmia Database (NIFEA DB) of 26 subjects to fine-tune the trained model to fine-tune the trained model based on active learning to detect anomalies in binary form for fetal. A variational autoencoder is trained on all data (adult and fetal ECG), and clustering is applied to latent features extracted from data after dimension reduction. Then, the sampling process of active learning selected samples from different clusters with low confidence to cover all data distribution. Moreover, a probability calibration based on mc-dropout and isotonic regression is used to calibrate confidences, helping to select reliable low-confidence samples. Various ablation studies were performed to show the influence of autoencoder-based sampling, calibration, and transfer learning, which showed that the proposed method could achieve 92% accuracy using 399 training samples. In contrast, other methods required more training samples to reach the same level of accuracy without calibration or an autoencoder and clustering approach or training without active learning. The study also found that transfer learning significantly impacted faster convergence and that the proposed active learning approach was more effective than traditional methods.
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 | 0.004 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.002 |
| 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".