Introduction: ============= This data provided in this repository contains supporting MATLAB results that underpin the figures and tables in the following manuscript: Title: An automated Quiet Sleep detection approach in premature infants as a gateway to assessing brain maturation Authors: A.Dereymaeker*; K.Pillay*; J.Vervisch; S.Van Huffel; G.Naulaers; K.Jansen; M.De Vos. Journal: International Journal of Neural Systems (IJNS) *These authors are joint first authors. Full citation details will be listed in the metadata for the repository when the paper becomes available. Data overview: ============== The following data are available in this repository: 1. A .zip folder 'CLASS_full.zip' that contains a .mat file of the CLASS algorithm results for each patient recording. This underpins the results in Figure 4 and part of Table 2. 2. A .zip folder 'CLASS_variants.zip' containing .mat files of all the CLASS algorithm variants results for each patient recording. This underpins the results in part of Table 2. 3. A .zip folder 'SATP.zip' containing the .mat file of the SAT% algorithm results for each patient recording. This underpins the results in part of Table 2. 4. Any of these .mat files also contain the list of anonymised patient recording IDs and their post-menstrual ages (PMAs) underpinning Figure 1. 5. A .zip folder 'maturational_trends.zip' that contains .mat files and a .csv file for all the extracted maturational characteristics of the patient recordings. Used for deriving the normal maturational trends underpinning the results Table 3. Data format: ============ All data is presented as .mat or .csv files. .mat files are compatable in versions of MATLAB R2014a onwards. Each patient recording is named using a unique identifier in the format _. Multiple recordings at different PMAs from the same patient have the the same patient number but different recording numbers. Full data desciption: ===================== Individual recording results: ----------------------------- The full list of results by the CLASS algorithm, for each recording in the test set, is located in the folder 'CLASS_full.zip' and named 'results_CLASS_test.mat'. This underpins the results of Figure 4 and part of Table 2. The .mat file contains two 2D cell matrices: 1. patient_results_full: This is a 56x15 cell matrix (including a header). Each row is an individual set of patient recording results. Each column shows the following: *ID: Patient ID in format _. *PMA: The postmenstrual ages of the recordings (whole weeks). *No. Clin. Events: The total number of identified Quiet Sleep (QS) 'events' as visually labelled by the clinicians. *No. CLASS Events: The total number of identified QS events as estimated by the CLASS algorithm. *The number of clinical QS events detected by CLASS *The number of clinical QS events NOT detected by CLASS *The number of CLASS estimated QS events aligned with clinical QS events. *The number of CLASS estimated QS events NOT aligned with clinical QS events. *Sensitivity between the clinical and CLASS-estimated QS signals. *Specificity between the clinical and CLASS-estimated QS signals. *Accuracy between the clinical and CLASS-estimated QS signals (not used in the final manuscript results). *A list of the QS durations (not calculated in this study so usually set to 0). *Proportion of the CLASS-estimate QS in the full recording (not used in the final manuscript results). *The sample binary signal of clinical QS events, with a sampling frequency of Fs=250/3 Hz. 0 denotes non-QS, 1 denotes QS. *The sample binary signal of CLASS-estimate QS events, with a sampling frequency of Fs=250/3 Hz. 0 denotes non-QS, 1 denotes QS. *Indicates with a 'Y' if a broken channel was removed from parts of the recording, during analysis. *F1 score between the clinical and CLASS-estimated QS signals (not used in the final manuscript results). Note: The included headers within the cell matrix use the reference 'TemP' which acts as a general placeholder to represent the results for the different versions of the CLASS algorithm, or the SAT% algorithm (described below). 2. patient_QS_curves: This is a 55x1 cell matrix. Each row aligns to the corresponding recording listed in the accompanying 'patient_results_full' file. This contains the QS detection signal before final thresholding is applied and was used for deriving the ROC plots (and AUC values) for the recordings. With the format as described above, the following .mat files are included for the various versions of CLASS and the SAT% algorithm that underpin the remaining results in Table 2. These can be found in the folders 'CLASS_variants.zip' and 'CLASS_SATP.zip' and named as follows: *results_CLASS_test_noASR.mat: The full results on the test set when CLASS is applied without ASR (bandpass filtering only). *results_CLASS_test_USG1.mat: The full results on the test set when CLASS is applied with uniform segmentation of 1 second. *results_CLASS_test_USG5.mat: The full results on the test set when CLASS is applied with uniform segmentation of 5 seconds. *results_CLASS_test_sd.mat: The full results on the test set when CLASS is applied with standard deviation alone (without multiple features and clustering). *results_CLASS_test_SATP.mat: The full results on the test set when the SAT% method is applied. Note: In some of the above .mat files, an additional two columns exist in 'patient_results_full'. These contain raw binary signals for clinician QS events and algorithm QS events, repectively, before correction for edge effects due to the moving average filters. These were not used in this study. Maturational characteristics: ----------------------------- All data for deriving the normal maturational curves presented in Table 3 can be found in the folder 'maturational_trends.zip'. The provided trend characteristics are calculated for the entire test set of recordings. A list of the the sub-set of recordings (age range 31-38 weeks PMA) and their corresponding 'continuous' PMAs (weeks and days) used to produce the normal maturational trends in Table 3, can be found in the file 'trend_ages.csv'. This .csv file is a 3-column list (with header) containing the following information: *patient number: The patient number of the recording *recording number: The corresponding recording number *PMA: The recording PMA in terms of weeks and days ('continuous' value). Note: patient numbers and recording numbers correspond to the earlier patient ID values which were in the form _. The full set of calculated characteristics, from the CLASS-estimated QS events in the test set, are located in the file 'CLASS_charac_test.mat'. This .mat file contains a 55x3 cell matrix called 'set_charac'. Each row is an individual set of patient recording results. Each column shows the following: *ID: Patient ID in format _. *PMA: The postmenstrual ages of the recordings (whole weeks). *7xN matrix of characteristics values. Each column is an nth extracted QS period from the recording. Each row denotes a calculated characteristic from the QS period. 7 raw characteristics are calculated in the following order: 1. Relative delta power. 2. Relative theta power. 3. Relative alpha power (not included in Table 3). 4. Relative beta power. 5. Full band power (not used in this study). 6. 90% spectral edge frequency (not used in this study). 7. Burst% Note: The logarithm of these characteristics are used for the trends produced in Table 3. With the format as described above, the following additional .mat files are included that underpin the remainder of the results in Table 3. These are named as follows: *visual_charac_test.mat: The maturational characteristics derived from clinician QS events. *EEG_charac_test.mat: The maturational characteristics derived from non-state specific EEG epochs.