Abstract 4366212: Artificial Intelligence with Augmented Data Increases Accuracy and Speed of Heart Murmur Detection in Phonocardiogram Data
Bibliographic record
Abstract
Introduction: Cardiac auscultation often provides the first indication of underlying cardiac conditions through identification of heart murmurs. Auscultation is conducted during routine physical exams and is traditionally performed via stethoscope. However, conventional auscultation has limited sensitivity and accuracy due to high inter-rater discrepancy. Given its importance in the diagnostic process, an enhanced protocol is necessary. Artificial intelligence (AI) has been implemented in various medical settings. However, AI is limited by its need for vast quantities of high-quality training data. Data augmentation (DA) can be used to generate new samples from existing data by applying various transformations, which increases algorithm robustness and generalizability. Hypothesis: We hypothesized that implementing the Minimally Random Convolutional Kernel Transform (MiniROCKET) AI model with DA techniques improves the accuracy and speed of murmur detection in phonocardiogram (PCG) data. Methods: We used data from the 2022 George Moody PhysioNet Heart Sound Classification Challenge, containing PCG recordings of individuals under 21 years of age in Northeast Brazil. Patients without recordings from all four heart valves were excluded. Each patient’s audio files were synchronized at the first heartbeat. The file group then underwent 10 random time-series DA techniques, yielding four files from each original. Then, a Mel spectrogram was generated from each file, and one random DA technique was applied to each to make three spectrograms (Figure 1). Through DA, our sample size was increased from 928 spectrograms to 14,848. After pre-processing, MiniROCKET classified each patient as: “murmur present,” “murmur absent,” or “further assessment required” - if sufficient certainty is not achieved. Results: We assessed the effect of varying levels of DA, the overall accuracy of our MiniROCKET methodology, and the speed of our model by comparing with existing models (Tables 1&2). Overall, our method yielded improved quality assessment metrics and enhanced speed compared to existing models and was further improved through DA techniques. Conclusion: With DA (p=8.600e-05), our method boasts rapid and precise detection of murmurs in PCG data, with a weighted accuracy of 96.4% and an evaluation time of 0.02 seconds per patient, outperforming existing methods. Implementing this method may streamline diagnosis, promote scalability and adaptability, and allow for early-stage treatment.
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 | 0.009 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 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".