@incollection{fdi:010095087,
  title = {{ECG}recover : a deep learning approach for electrocardiogram signal completion},
  author = {{L}ence, {A}. and {G}ranese, {F}. and {F}all, {A}hmad and {H}anczar, {B}. and {S}alem, {J}.{E}. and {Z}ucker, {J}ean-{D}aniel and {P}rifti, {E}di},
  editor = {},
  language = {{ENG}},
  abstract = {{I}n this work, we address the challenge of reconstructing the complete 12-lead {ECG} signal from its incomplete parts. {W}e focus on two main scenarios: (i) reconstructing missing signal segments within an {ECG} lead and (ii) recovering entire leads from signal in another unique lead. {T}wo emerging clinical applications emphasize the relevance of our work. {T}he first is the increasing need to digitize paper-stored {ECG}s for utilization in {AI}-based applications, often limited to digital 12 lead 10s {ECG}s. {T}he second is the widespread use of wearable devices that record {ECG}s but typically capture only one or a few leads. {I}n both cases, a non-negligible amount of information is lost or not recorded. {O}ur approach aims to recover this missing signal. {W}e propose {ECG}recover, a {U}-{N}et neural network model trained on a novel composite objective function to address the reconstruction problem. {T}his function incorporates both spatial and temporal features of the {ECG} by combining the distance in amplitude and sycnhronization through time between the reconstructed and the real digital signals. {W}e used real-life {ECG} datasets and through comprehensive assessments compared {ECG}recover with three state-of-the-art methods based on generative adversarial networks ({EKGAN}, {P}ix2{P}ix) as well as the {C}opy{P}aste strategy. {T}he results demonstrated that {ECG}recover consistently outperformed state-of-the-art methods in standard distortion metrics as well as in preserving critical {ECG} characteristics, particularly the {P}, {QRS}, and {T} wave coordinates.},
  keywords = {},
  booktitle = {{P}roceedings of the 31st {ACM} {SIGKDD} {C}onference on {K}nowledge {D}iscovery and {D}ata {M}ining {V}.1},
  numero = {},
  pages = {2359--2370},
  address = {{N}ew {Y}ork},
  publisher = {{A}ssociation for {C}omputing {M}achinery},
  series = {},
  year = {2025},
  DOI = {10.1145/3690624.3709405},
  ISBN = {979-8-4007-1245-6},
  URL = {https://www.documentation.ird.fr/hor/fdi:010095087},
}