Fig 1.
Pipeline for ARGO construction and exclusion criteria. Flow diagram illustrating the selection and filtering process used to build the ARGO dataset.
Table 1.
Number of EGMs annotated for each patient and included in the ARGO dataset.
Table 2.
Distribution of EGM categories across the nine patients obtained with the consensus. The table reports the number of annotated AVPs, physiological and unknown EGMs for each patient, along with the total number of EGMs per patient and across the entire dataset.
Table 3.
Summary of reliability in class annotation. Fleiss’ κ was used for overall inter-annotator and intra-annotator reliability and Cohen’s κ for individual versus consensus reliability.
Fig 2.
Contingency tables comparing the class annotations independently provided by each rater against the class annotations defined during the consensus meetings.
From left to right, the contingency tables that compare the classification of Annotator1, Annotator2 and Annotator3 against the consensus across the categories AVP (A), Physiological (P) and Unknown (U). In each table, the number of EGMs that were classified coherently by both the annotator and the consensus (out of the 1962 labelled EGMs) was highlighted in bold.
Table 4.
Summary of reliability in AVP delineation. ICC(3,1) was employed to assess inter-rater, intra-rater, and individual raters against consensus agreement.
Fig 3.
Distribution of bipolar EGM peak-to-peak amplitudes across annotated classes.
From left to right, violin plots of bipolar EGM peak-to-peak amplitudes for EGMs annotated as Physiological (green), AVP (red), and Unknown (yellow).
Fig 4.
Distribution of AVP duration of annotated AVPs.