Atrial fibrillation (AF) is the most common sustained cardiac arrhythmia seen in the clinical practice affecting 1-2% general population. Catheter ablation is an effective treatment for early stage of the disease (paroxysmal AF) by inserting catheters in the heart to ‘burn’ (ablate) to eliminate the triggers causing AF. However, results are poor in more advanced form of the disease (persistent AF) as there is interplay between trigger and substrate whereby simply eliminating the triggers would not produce an effective outcome. Accurate identification of relevant substrate targets cannot be achieved currently as the electrical signals (electrograms) during AF exhibit complex and chaotic behaviour which requires more advanced methods of analysis. A variety of signal processing techniques (i.e. complex fractionated electrogram, dominant frequency, local activation time mapping and rotational focal activity by phase mapping) have been developed and applied in clinical studied but failed to generate consistent acceptable outcome for AF treatment. Usually, a certain set of ‘rules’ were pre-defined for such single-hypothesis-driven techniques. However, multiple mechanisms may co-exist during AF.
Novel machine learning algorithms1 based on advanced signal processing techniques2-4 can provide better performance than traditional methods, in unveiling the underlying mechanisms of persistent atrial fibrillation and provide new insight in the strategy of catheter ablation in treating persistent atrial fibrillation.
This project will develop a bespoke high precision diagnostic tool for guiding AF ablation, combining novel machine learning (ML) architectures and advanced signal processing techniques. This research project is highly interdisciplinary, bringing together the expertise from cardiac electrophysiology, biomedical engineering and artificial intelligence involving researchers at Leicester and internationally.
The following specific aspects are the key activities of the proposed research project:
1. Improve the performance of ML model1 and capture patient-specific characteristics using handcrafted features developed at Leicester from a large cohort of patients. The post-holder will develop ML tools based on the large database of clinical data already collected with atrial electrograms on the response to ablation for the termination of AF. This will lead to the development of new ML classification models.
2. To develop ML models that will combine the best handcrafted features2-4 using both labelled and unlabelled data; investigate the effect of handcrafted features; compare between handcrafted features and automatic generated features
3. To identify the best ML model for the stratification of AF patients. The selected ML model will be embedded in the final platform solution that will be used in prospective studies to treat AF.
4. A graphic interface integrating the trained model will be developed to enable real-time analysis to support a prospective clinical study.
This project will provide novel and unique insights into the understanding of the underlying mechanisms using machine learning methods based on large-scale database, which has great potential to facilitate industrial collaboration and further translation into clinical tools for effective interventional treatment and improved patient outcome.
1. Li X, Chu GS, Almeida TP, Salinet JL, Mistry AR, Vali Z, Stafford PJ, Schlindwein FS and Ng GA. A k-Nearest Neighbours Classifier for Predicting Catheter Ablation Responses Using Noncontact Electrograms During Persistent Atrial Fibrillation. 2018 Computing in Cardiology Conference (CinC). 2018;45:1-4.
2. Li, X., Chu, G.S, Almeida, T.P., Vanheusden F.J., Salinet, J., Dastagir, N., Mistry, A. R., Vali, Z, Sidhu B, Starfford, P. J., Schlindwein, F.S., Ng, G.A., 2021. Automatic Extraction of Recurrent Patterns of High Dominant Frequency Mapping during Human Persistent Atrial Fibrillation. Front. Physiol, 2021 Vol. 12 Issue 286, DOI: 10.3389/fphys.2021.649486
3. Li, X., Almeida, T.P., Dastagir, N., Guillem, M.S., Salinet, J., Chu, G.S., Stafford, P.J., Schlindwein, F.S., Ng, G.A., 2020. Standardizing Single-Frame Phase Singularity Identification Algorithms and Parameters in Phase Mapping During Human Atrial Fibrillation. Front. Physiol. 11, 869. doi: doi.org/10.3389/fphys.2020.00869
4. Almeida TP, Schlindwein FS, Salinet J, Li X, Chu GS, Tuan JH, et al. Characterization of human persistent atrial fibrillation electrograms using recurrence quantification analysis. Chaos. 2018;28(8).
5. Li X, Salinet JL, Almeida TP, Vanheusden FJ, Chu GS, Ng GA, et al. An interactive platform to guide catheter ablation in human persistent atrial fibrillation using dominant frequency, organization and phase mapping. Computer Methods and Programs in Biomedicine. 2017;141:83-92.