This README.txt file was generated on 2018-06-06 by Nadine Chang ------------------- GENERAL INFORMATION ------------------- 1. Title of Dataset: BOLD5000: Brain, Objects, Landscapes Dataset 2. Author Information First Author Contact Information Name: Nadine Chang Institution: Carnegie Mellon University Address: 5000 Forbes Ave., Pittsburgh, PA 15213 Email: nchang1@cs.cmu.edu Second Author Contact Information Name: John Pyles Institution: Carnegie Mellon University Address: 5000 Forbes Ave., Pittsburgh, PA 15213 Email: jpyles@cmu.edu Third Author Contact Information Name: Abhinav Gupta Institution: Carnegie Mellon University Address: 5000 Forbes Ave., Pittsburgh, PA 15213 Email: abhinavg@cs.cmu.edu Fourth Author Contact Information Name: Michael Tarr Institution: Carnegie Mellon University Address: 5000 Forbes Ave., Pittsburgh, PA 15213 Email: michaeltarr@cmu.edu Last Author Contact Information Name: Elissa Aminoff Institution: Fordham University Address: 441 E Fordham Rd., Bronx, NY, 10458 Email: eaminoff@fordham.edu 3. Date of fMRI collection: 2017-10-27 to 2018-02-14 4. Geographic location of data collection: Pittsburgh, PA, USA 5. Information about funding sources that supported the collection of the data: 6. Data collection is supported by NSF Award BCS-1439237 to Elissa Aminoff & Michael Tarr and also by ONR MURI N000141612007, Sloan, Okawa Fellowship to Abhinav Gupta. -------------------------- SHARING/ACCESS INFORMATION -------------------------- 1. Licenses/restrictions placed on the data: CC0 2. Please refer to our website for further information and updates on new data releases. BOLD5000.org 3. For data usage and citation, please refer to BOLD5000.org. --------------------- DATA & FILE OVERVIEW --------------------- Directory of Files 1. Filename: BOLD5000_Scan_Information.txt Short description: Scanning details for any abnormal events during data collection. 2. Filename: BOLD5000_Subject_Information.zip Short description: Contains all relevant subject information. Contains Daily_Intake folder for all subjects' daily intakes. Contains Subject_Demographics.txt with subject demographics. 3. Filename: BOLD5000_Stimuli.zip Short description: Contains all relevant stimuli information. Contains Image_Labels folder with all image labels for all stimuli. Contains Localizer_Stimuli folder with all stimuli used for the functional localizer. Contains Scene_Stimuli folder with the presented stimuli and the original uncropped images corresponding to the stimuli. Also contains a list of the repeated stimuli. Contains Stimuli_Presentation_List folder with the stimuli presentation order for each subject for each session for each run. 4. Filename: BOLD5000_CSIX_sessYY.zip (X is subject number, YY is session number) Short description: Contains all data collected in that subject's session. Contains Behavioral_Data folder with all behavioral data. Contains BOLD_Raw folder with all raw BOLD data. Contains physio folder with all physiological data. 5. Filename: BOLD5000_Unfiltered_CSIX_sessYY.zip (X is subject number, YY is session number) Short description: Contains all unfiltered (no Siemens Prescan Normalization filter) raw BOLD data collected in that subject's session. 6. Filename: BOLD5000_Structural.zip Short description: Contains all subject's structural raw MRI data. 7. Filename: BOLD5000_ROIs.zip Short description: Contains all subject's processed ROIs voxels in vector form. File naming convention: Subject identifiers are CSI1, CSI2, CSI3, CSI4 ------------------------------------------------ DATA DESCRIPTION FOR: coco_final_annotations.pkl ------------------------------------------------ This file contains a python dictionary for all annotations for each of the COCO images used as stimuli in this study. Note that the annotations are for the original COCO images of the later cropped images used for stimuli. The dictionary of annotations are contained in a pickle file, and the annotations are kept in the same organization as given by the COCO dataset. In this pickle file, the main dictionary has COCO image ids(the numbers in the image name) as keys and a list of its corresponding image annotations as values. Each element in the annotation list is a detailed dictionary for one annotation and contains following keys: [u'segmentation', u'area', u'iscrowd', u'image_id', u'bbox', u'category_id', u'id’] example: for image COCO_train2014_000000580813.jpg image id = 580813 annotations[580813] will load a list of annotations for this image Please refer to the COCO website for further details. On the website there is also a publicly available API to load and draw annotations. cocodataset.org ----------------------------------------------- DATA DESCRIPTION FOR: imagenet_final_labels.txt ----------------------------------------------- This file contains the category names that correspond to each synset in ImageNet. The synsets are in the image name. example: for image n15075141_17963.JPEG synset = n15075141 a search in the file will show that synset n15075141 corresponds to toilet tissue, toilet paper, bathroom tissue. -------------------------------------------- DATA DESCRIPTION FOR: scene_final_labels.txt -------------------------------------------- This file contains the category names for all scene images. The image name contains the category name. example: for image yogastudio4.jpg category = yoga studio ---------------------------------------------- DATA DESCRIPTION FOR: BOLD5000_CSIX_sessYY.zip ---------------------------------------------- - DICOM_log_XXXXXXXXXXX.txt is a log file showing all scans collected during the session. The long number identifier in the .txt filename is the unique identifier for each session used by the imaging center. This is also the identifier in the DICOM file header. - Two series with the same name appear in the DICOM_log_XXXXXXX.txt file due to both filtered and un-filtered data being saved. (First series is unfiltered, second is filtered with Siemens Prescan Normalization.) - Subject age is recorded in the DICOM header as 01/01/YEAR where YEAR is the year of the subject’s birth. Day and month are removed for privacy. - The “localizer” and “Trufi Sag” scans listed in the DICOM_log_XXXXXXX.txt were not included in the dataset since they are scout scans. - The “fieldmap” scans were not included in the dataset due to a technical error with sub-optimal parameters. - During the slow-event related scene functional runs, subjects received the following instructions before each run: “Please keep your eyes focused on the middle of the screen at all times. You will see pictures of scenes and objects. Please press the right index key for like, middle for neutral, ring finger for dislike. Please respond after the image is shown.” In the .tsv files in the Behavioral_Data folder coding for the “Response” column is as follows: 1 = like 2 = neutral 3 = dislike - For BOLD runs denoted as “SceneLocal” and behavioral data denoted as “localizer” these are the functional localizer runs used to define category selective regions using three stimuli categories: Scenes, Objects, and phase scrambled scenes. Subjects’ performed a one-back task. ---------------------------------------------- DATA DESCRIPTION FOR: BOLD5000_Structural.zip ---------------------------------------------- - Structural scans that include a T1-MPRAGE, T2-SPACE, and a 230 direction diffusion (DSI) scan. ---------------------------------------------- DATA DESCRIPTION FOR: BOLD5000_ROIs.zip ---------------------------------------------- - CSIX_ROIs_TRY.h5 or .mat contains processed ROIs voxels. ROIs provided are 'LHEarlyVis', 'RHEarlyVis', 'LHLOC', 'RHLOC', 'LHPPA', 'RHPPA', 'LHRSC', 'RHRSC', 'LHOPA', 'RHOPA’. Each ROI voxels come in matrix form. The matrix shape is # of images x # of dimensions for that ROI. Note that for CSI1, 2, 3 the # of images viewed is 5254. For CSI4 the # of images viewed is 3108 example: For CSI1 ‘LHPPA’ Matrix is of shape 5254 x 157 The ordering of images are provided in stim_list folder. CSIX_stim_lists.txt provided the 5254 ordering of image names. -------------------------- METHODOLOGICAL INFORMATION -------------------------- MRI data was acquired on a 3T Siemens Verio scanner at the Scientific Imaging and Brain Research Center on the Carnegie Mellon University using a 32-channel head coil. BOLD fMRI Functional images were collected using a T2*-weighted echoplanar imaging pulse sequence. -69 slices parallel to the AC/PC -In-plane resolution = 2mm x 2mm -Slice thickness = 2mm (no gap) -Field of View = 212mm -Matrix Size = 106 -Interleaved acquisition -6/8th phase partial Fourier -Multi-band factor = 3 (CMRR multi-band sequences) -TR = 2000ms -TE = 30ms -Flip angle = 79degrees -Advanced Shim Mode Note: BOLD EPI data was collected using Siemens Prescan Normalization filter. As some researchers prefer unfiltered images, we also saved our data without Prescan Normalization which can be downloaded in BOLD5000_Unfiltered_CSIX_sessYY.zip files. Each scene run contained 194 volumes, and each functional localizer run contained 141 volumes. OppPhase-BOLD_PA_CSIX-X To allow distortion correction using an opposite phase encoding approach in each session a 3 volume run was acquired using the “Invert RO/PE Polarity” option with parameters exactly matching the BOLD fMRI parameters except for PA phase encoding. SPIN-ECHO FIELDMAP To provide another option for distortion correction, we acquired two pairs of spin-echo scans with opposite phase encoding. The primary difference between these two versions is the use of partial Fourier which results in different contrast. Researchers are encouraged to use the version they find provides the best results. SpinEchoFieldMap_pF68_AP_CSIX_Sess-X -69 slices parallel to the AC/PC -In-plane resolution = 2mm x 2mm -Slice thickness = 2mm (no gap) -Field of View = 212mm -Matrix Size = 106 -6/8th phase partial Fourier -TR = 6708ms -TE = 45ms -Flip angle = 90degrees SpinEchoFieldMap_AP_CSIX_Sess-X -69 slices parallel to the AC/PC -In-plane resolution = 2mm x 2mm -Slice thickness = 2mm (no gap) -Field of View = 212mm -Matrix Size = 106 -TR = 9240ms -TE = 81.6ms -Flip angle = 90degrees STRUCTURAL SCANS Structural scans were acquired across 2-3 sessions separate from the fMRI sessions. T1w_MPRAGE -176 sagital slices -1x1x1mm resolution -Matrix size = 256 -TR = 2300ms -TE = 1.97ms -TI = 900ms -Flip angle = 9degrees -GRAPPA factor = 2 T2wSPACE -176 sagital slices -1x1x1mm resolution -Matrix size = 256 -TR = 3000ms -TE = 422ms -TI = 900ms -GRAPPA factor = 2 Subjects’ faces were removed to protect privacy and converted to NIfTI format DSI_MB3-253d_AP -Diffusion spectrum imaging sampling scheme -69 slices parallel to the AC/PC -In-plane resolution = 2mm x 2mm -Slice thickness = 2mm (no gap) -Field of View = 212mm -Matrix Size = 106 -6/8th phase partial Fourier -TR = 3981ms -TE = 121ms -Direction = 230 -Maximum b-value = 3980 s/mm2 -Bipolar acquisition DSI_MB3-253d_PA -Same as above but with PA phase encoding (achieved using the “Invert RO/PE Polarity” option) To reduce motion and maintain consistent head placement and alignment across sessions, Headcases (CaseForge, Inc.) customized for each participant were used.