TARGET-MAP: Transcranial doppler AutoReGulation assessment of brain perfusion in nEurocritical patients for MAP optimization

@article{PhysioNet-target-map-1.0.0,
  author = {{Trombin Marques}, Marina and Bao, Kevin and Hayman, Adeline and Peterson, Katherine and Lee, Eliot and Vitt, Jeffrey and Hu, Xiao and smith, wade and Hemphill, Claude and {Sampaio Silva}, Gisele and Amorim, Edilberto},
  title = {{TARGET-MAP: Transcranial doppler AutoReGulation assessment of brain perfusion in nEurocritical patients for MAP optimization}},
  journal = {{PhysioNet}},
  year = {2026},
  month = oct,
  note = {Version 1.0.0},
  doi = {10.13026/b53c-b825},
  url = {https://doi.org/10.13026/b53c-b825}
}

@article{pollard_physionet_2026, title = {{PhysioNet} as a global platform for biomedical research}, volume = {1}, issn = {3005-0693}, url = {https://doi.org/10.1038/s44360-026-00096-z}, doi = {10.1038/s44360-026-00096-z}, abstract = {Twenty-five years after its first public repository was established, the PhysioNet resource for biomedical and clinical data has enabled tens of thousands of research papers. Further efforts are ongoing to advance secure data sharing, community engagement and scalable research infrastructure.}, number = {8}, journal = {Nature Health}, author = {Pollard, Tom and Moody, Benjamin E. and Lehman, Li-wei H. and Gow, Brian J. and Fernandes, Chrystinne and Xie, Chen and Johnson, Alistair and Mark, Roger G. and Heldt, Thomas}, month = aug, year = {2026}, pages = {792--795},}

Abstract

The Transcranial doppler AutoReGulation assessment of brain perfusion in nEurocritical patients for MAP optimization (TARGET-MAP) Database includes baseline clinical information, continuous transcranial doppler (TCD), and data from multiparameter monitor (blood pressure, electrocardiogram, end-tidal CO2 [ETCO2], and SpO2) monitoring from 24 neurocritical patients from an academic center in the USA. The monitoring sessions are 30-60 minutes long and were recorded for up to 5 days after patients' admission in the intensive care unit (ICU). The neurological outcome was determined using the Cerebral Performance Category scale at discharge.

Background

Following the primary brain injury, a cascade of maladaptive physiologic processes determines brain edema, seizures, metabolic failure, and impaired cerebrovascular reactivity, leading to secondary brain injury. [1] Preventing these mechanisms is a key strategy when managing acute brain injury patients. Cerebral autoregulation is a homeostatic process responsible for maintaining a constant cerebral blood flow across a range of blood pressures. Neurocritical patients, such as traumatic brain injury (TBI), post-brain injury, and stroke patients, often present impaired autoregulation, which makes them prone to ischemia. [2]

Multimodal point-of-care non-invasive CBF monitoring, utilizing simultaneous continuous TCD and EEG, has the potential to facilitate the detection of brain ischemia and prevent brain injury. [3,4] Knowledge from this work will determine whether TCD’s non-invasive autoregulation metrics can be used to individualize specific mean arterial pressure (MAP) thresholds and if EEG may be a suitable target for MAP titration.

Methods

The database originates from an academic center in the U.S. (Zuckerberg San Francisco General Hospital, San Francisco, CA, USA)

This database comprises clinical, transcranial Doppler (TCD), and vital signs monitoring data from adult patients with neurocritical conditions in a neurointensive care unit.

Monitoring was typically started within 48 hours of an acute brain injury and was performed up to once a day in the first five days from injury. The study duration was approximately 1 hour, depending on the patient's stability and availability for TCD monitoring. In addition to TCD data and arterial blood pressure data, this database may include EEG or ECG data.

Clinical Data

Patient information was recorded at the time of admission (age and sex). In patients with cardiac arrest cases, we collected the location of arrest (out or in-hospital), type of cardiac rhythm recorded at the time of resuscitation (shockable rhythms include ventricular fibrillation or ventricular tachycardia and non-shockable rhythms include asystole and pulseless electrical activity), the time between cardiac arrest and ROSC, and temperature control, and vasopressors during each monitoring session.

Monitoring data

TCD data was collected using a Multigon robotic device (Multigon ROBOTOC9MD). Vital signs were collected through the Philips IntelliVue MX800 monitor. Signals were integrated and recorded using a Moberg CNS 200. The Moberg monitor receives and integrates different time series from various monitors; however, it cannot synchronize the internal clocks of these devices. Hence, there may be an inconsistent delay of approximately 1 second between TCD data from the Multigon and the vital signs from the IntelliVue monitor.

The HDF5 file contains the original recording, without synchronization of the signals across waveform sources. The CSV files contain ABP and TCD signals synchronized using the Haemosync algorithm, which applies a constant delay followed by a cross-correlation synchronization. [5]

Waveform signals may be stored as integers or floats. For integer waveforms, raw values are rescaled to physical units using a linear transformation: Unitized value = minMV + ((rawValue − minSV) × (maxMV − minMV) / (maxSV − minSV)). In our dataset, ECG signals are sampled at 500 Hz, whereas ABP, TCD, SpO2, ETCO2 are sampled at 125 Hz. Each waveform has an associated “,settings” file that specifies the storage range (minSV, maxSV), the physical measurement range (minMV, maxMV), and the measurement units.

Patient Outcomes

In cardiac arrest patients, clinical outcome was determined using the best Cerebral Performance Category (CPC) scale at discharge. CPC is an ordinal scale ranging from 1 (good neurological function) to 5 (death).

De-identification

Clinical and monitoring data were de-identified. Patients aged 89 years or older are listed with age "90".

Data Description

Anthropic Claude 4.5 Haiku Thoughts Copy thoughts to clipboard

There are two folders, called HDF5_Trimmed and CSV only, both of which contain de-identified monitoring data (Table 1 describes the filenames, formats, and sizes).

The CSV_only folder contains the columns Time in seconds, starting from 0 in each recording; ABP (arterial blood pressure) in mmHg; and LMCA (left middle cerebral artery mean velocity) and RMCA (right middle cerebral artery mean velocity) in cm/s. Some patients had unilateral monitoring for some blood vessels due to a limited temporal bone window, If unilateral PCA monitoring was pursued, some files may only include one side monitoring (e.g., LPCA [left posterior cerebral artery mean velocity] or RPCA [right posterior cerebral artery mean velocity]), and it is signalized on its name. The timestamps start at 0.

The HDF5_Trimmed contains the raw, deidentified, complete monitoring data, not synchronized. It is named after the research participant ID (TAR001, TAR002, and so on) and according to the day of the recording (D1, D2, D3, D4 or D5). The contents of each HDF5 file can be checked using the following MATLAB script [5].

The TCD-derived metrics available in the HDF5 files are:

• Peak: systolic peak velocities
• Dias: diastolic velocity
• Mean: mean velocity
• RI: resistance index
• PI: pulsatility index
• Depth: depth of insonation
• Gate: gate size (or sample volume)
• AMPL: amplitude
• SD: systolic/diastolic
• SBI: Spectral Broadening Index

Dataset structure: /Signal_Type,Channel_or_Label,DataStructure,DataType,Device

• Signal_Type describes the type of data, e.g., TCD, ABP, ECG.
• Channel_or_Label identifies the channel providing the signal, for instance, LMCA-U1 means left middle cerebral artery coming from channel U1.
• DataStructure provides information about the sort of signal, when "SampleSeries" it refers to the TCD waveforms; while if it is, e.g., Mean, PI, RI it refers to computed metrics based on TCD.
• DataType reflects numeric representation, e.g., Integer or Float (decimal numbers).
• Device refers to the source hardware, in this case, IntelliVue is the Philips patient monitor and Multigon is the TCD machine.

Example of contents inside a HDF5 file:

📁 Group:
📊 Dataset: /TCD,LMCA-L1,AMPL,Float,Multigon
📊 Dataset: /TCD,LMCA-L1,AMPL,Float,Multigon,timestamps
📊 Dataset: /TCD,LMCA-L1,Depth,Float,Multigon
📊 Dataset: /TCD,LMCA-L1,Depth,Float,Multigon,timestamps
📊 Dataset: /TCD,LMCA-L1,Dias,Float,Multigon
📊 Dataset: /TCD,LMCA-L1,Dias,Float,Multigon,timestamps ...

📁 Group: //Waveforms
📊 Dataset: //Waveforms/ABP,na,SampleSeries,Integer,IntelliVue
📊Dataset: //Waveforms/ABP,na,SampleSeries,Integer,IntelliVue,timestamps
📊 Dataset: //Waveforms/ECG,II,SampleSeries,Integer,IntelliVue
📊 Dataset: //Waveforms/ECG,II,SampleSeries,Integer,IntelliVue,timestamps
📊 Dataset: //Waveforms/TCD,LMCA-U1,SampleSeries,Integer,Multigon
📊 Dataset: //Waveforms/TCD,LMCA-U1,SampleSeries,Integer,Multigon,timestamps

The HDF5 file is organized hierarchically. The root group contains derived Transcranial Doppler (TCD) parameters for bilateral middle cerebral arteries at multiple depths, including systolic, diastolic, mean velocities, and pulsatility indices, each with associated timestamps. The /Trends group stores low-frequency numeric trends of vital signs from the bedside monitor (arterial blood pressure, heart rate, respiratory rate, and oxygen saturation). The /Waveforms group contains high-frequency raw physiological signals, including arterial blood pressure, ECG, respiratory waveforms, and raw TCD velocity signals, all timestamped for synchronization across modalities.

The deidentified_target_clinical_data_final.csv contains the de-identified clinical data for all enrolled cardiac arrest patients.

Table 1. Filenames, format and sizes

Usage Notes

Please refer to these instructions [6] to utilize the code and explore the contents of the HDF5 files [7].

The intended use of this dataset is to analyze cerebral perfusion and autoregulation using Transcranial Doppler. It is important to note that there is a delay between TCD and vital signs waveforms recordings due to the fact that they are obtained from separate devices that cannot have clock synchronized directly. Therefore, waveforms may need to be synchronized depending on the analysis planned. The signals in both HDF5 and CSV files must be inspected and cleaned according to the analysis plan.

In the HDF5 file, timestamps are in microseconds, elapsed since 1970-01-01 00:00:00 UTC. The waveform deidentification involved date-shifting by a random number of days (from 1 to 365 days). There is also conversion from local time (PST) to UTC.

The CSV file deidentified_target_clinical_data_final.csv contains de-identified clinical data for cardiac arrest patients; the columns 1st–5th_TCD_diff_from_ROSC indicate the time elapsed from ROSC to each monitoring session.

Ethics

The authors declare no ethical concerns. This project was approved by the Institutional Review Board (IRB), under the IRB number 24-41684.

Informed consent was pursued at the moment of enrollment with a surrogate. In the absence of a surrogate, delayed consent was pursued once legal authorized representatives (LAR) were identified. In case of patients who had no LAR, need for consent was waived.

Clinical and monitoring data were de-identified, not containing name, date, time, or any personal health information (PHI). Patients aged 89 years or older are listed with age "90". The identifiers do not identify the hospital name.

Acknowledgements

We gratefully acknowledge the dedicated contributions of the nursing staff, respiratory therapists, occupational and physical therapists, EEG technologists, radiology technicians, and the broader multidisciplinary support staff at Zuckerberg San Francisco General Hospital. Their expertise, professionalism, and unwavering commitment to patient care were essential to both the care of our patients and the successful completion of this work.

Conflicts of Interest

Dr. Marques has received support from the American Heart Association (24DIVSUP1274116) to develop this project.

Dr. Amorim has received support from the NIH (K23NS119794, 1OT2OD032701, R01NS128342), Department of Defense (HT9425-23-1-0242, HT9425-25-1-0170, W81XWH-19-1-0861, W81XWH-21-C-0075), American Heart Association (AMFDP Faculty Development Program [843457], 20CDA35310297, 24DIVSUP1274116), Regents of the University of California, Cures Within Reach (2022CAL-Amorim), Zoll Foundation, and Hellman Foundation.

References

1. Sekhon MS, Ainslie PN, Griesdale DE. Clinical pathophysiology of hypoxic ischemic brain injury after cardiac arrest: a “two-hit” model. Crit Care. 2017 Apr 13;21(1):90.
2. Ameloot K, Genbrugge C, Meex I, Jans F, Boer W, Vander Laenen M, et al. An observational near-infrared spectroscopy study on cerebral autoregulation in post-cardiac arrest patients: Time to drop ‘one-size-fits-all’ hemodynamic targets? Resuscitation. 2015 May;90:121–6.
3. Zeiler FA, Czosnyka M, Smielewski P. Optimal cerebral perfusion pressure via transcranial Doppler in TBI: application of robotic technology. Acta Neurochir (Wien). 2018 Nov;160(11):2149–57.
4. Chen HY, Elmer J, Zafar SF, Ghanta M, Moura Junior V, Rosenthal ES, et al. Combining Transcranial Doppler and EEG Data to Predict Delayed Cerebral Ischemia After Subarachnoid Hemorrhage. Neurology [Internet]. 2022 Feb [cited 2024 July 18];98(5). Available from: https://www.neurology.org/doi/10.1212/WNL.0000000000013126
5. Eleveld N, Harmsen M, Elting JWJ, Maurits NM. Haemosync: A synchronisation algorithm for multimodal haemodynamic signals. Comput Methods Programs Biomed. 2024 Sept;254:108298.
6. https://github.com/edilbertoamorim/TARGETMAP/blob/main/README.md
7. https://github.com/edilbertoamorim/TARGETMAP/blob/main/Check_content_hdf5.m