ac01e21bb14609416760f790dd8966ae-Supplemental-Datasets_and_Benchmarks.pdf

In the hospital, patients may be in the ICU with ECG/PPG sensors to monitor their already-poor healthcondition. ML methods must rely onlearning toimpute missing signals based onthesignal that is present, rather than learning tocreate ageneral-purpose imputation template thatmimics standard healthybehavior. Likewise, participant movement inboth contexts can result in artifacts(e.g. Inabroadercontext, we want to match the high quality level of other datasets such as PTB-XL, in which 77.01% of thesignal data areofhighest assessed quality [18]. See below for examples of ECG signals with their associated periodogram.