HITI Lab
DatasetsEPIP

Overview

Overview of the Emory Prostate Imaging Project (EPIP)

This dataset is currently in-development so check back soon for updates!

Summary

The sample of the Emory Prostate Imaging Project (also known as EPIP) contains 844 prostate MRI exams for 840 patients over the period of 1.5 years.

Primer on Prostate Magnetic Resonance Imaging (MRI)

A full discussion on screening prostate MRI is beyond the scope of this documentation, however we will provide a brief primer. Prostate cancer is uncontrolled growth of cells in the prostate gland. It is the most common non-skin cancer in men in the United States, and the 2nd most common cancer in men worldwide. Individualized strategies for screening and management of localized prostate cancer are needed to prevent progression while also avoiding overtreatment.

In the United States, males are recommended Prostate-Specific Antigen (PSA) testing every 2 to 4 years between the ages of 50 and 69, with individualized strategies for screening for high-risk patients. Decisions on how to move forward with testing, imaging, biopsy and treatment are made in conjunction with clinical decision making.

Following a screening discussion which may include discussions about family history of prostate cancer, symptoms, current medications, and other risk factors, as well as a digital rectal exam (DRE), a patient may be recommended for PSA testing. A patient with an elevated PSA result, returns for retesting of PSA levels after 4-8 weeks. If levels remain elevated, a patient will proceed with Magnetic Resonance Imaging (MRI). If a suspicious lesion is identified (often PI-RADS 3-5) in the MRI, often, a biopsy is performed. The biopsy may be guided by ultrasound, MRI or a fusion of both methods. The biopsy results categorize the lesions as benign or grade them by how aggressive the prostate cancer cells appear under a microscope (International Society of Urological Pathology Grade). After cancer is confirmed and graded, PSMA PET can be used to for prostate cancer staging.

Note: PSMA PET may be used as complementary imaging when MRI does not give a clear result.

Summary Statistics

  1. Total Patients: 840
  2. Total Exams: 844
  3. Total Series: 3,383
  4. Manufacturers:
    1. Siemens: 93.44%
    2. GE: 6.56%
  5. Race:
    1. African American: 40.95%
    2. Asian: 3.81%
    3. White: 48.21%
  6. Ethnicity: 2.98% Hispanic

File Structure

Images in the dataset are structured as /patient_ID/study_ID/series_ID/DICOM

EPIP consists of one image modality and six primary tables that are used to assign clinical factors and image metadata.

  1. Exam contains exam-level information. Information includes patient demographics, encounter metadata, radiology report, diagnosis, prostate volume, PSA, prostate density, PCa history, number of lesions, prior imaging orders, and following image orders.
  2. Lesion contains lesion information. Information including lesion description, size, laterality, zone and PI-RADS score.
  3. Pathology contains specimen information. Information includes collection date, Gleason score, ISUP grade group, and diagnosis.
  4. Staging_spread contains information on tumor staging, invasion and metastasis.
  5. Treatment contains information about surgery, radiation and medication.
  6. Metadata contains image level information and is structured as one row per file. Information includes DICOM metadata, series description, and file path.

Exam Data

Exam contains all clinical data and demographics data collected for an exam. Each row will contain information regarding one exam.

Data Types

  1. Findings/descriptors - generated by the radiologist at the time of interpretation and extracted in a structured format by the HITI lab. Includes information about the prostate dimensions and the radiologist's narrative.
  2. PSMA - includes information about PSMA PET procedures prior to and following the exam.
  3. Global information - this includes study descriptions, and patient demographics.

Key fields

Feature nameTypeDescription
empi_anonstringUnique anonymized patient ID, all exams for a patient will have the same ID
accession_numstringUnique accession number of imaging exam
pat_age_at_examintegerPatient age at exam time
race_descstringpatient race
ethnic_group_descstringpatient ethnicity
begin_exam_dttmdatetimeExam start timestamp
proc_namestringName of imaging procedure
ord_clin_indstringClinical indication text from order
first_dxstringFirst prostate cancer diagnosis noted in EMR
diagnosisstringPrimary diagnosis for this exam
regionstringImaging location or region
has_cancer_during_exambooleanCancer present/confirmed during exam
prior_psma_datedateprior PSMA imaging date
prior_psma_proc_namestringprior PSMA imaging procedure name
prior_psma_impressionstringprior PSMA report impression
following_psma_datedatenext PSMA imaging date
following_psma_proc_namestringnext PSMA imaging procedure name
following_psma_impressionstringnext PSMA imaging report impression
volumefloatProstate gland volume (cc)
psa_densityfloatProstate-specific antigen level
clinical_indication_scorestringClinical indication score
historystringClinical history
extracted_lesion_nointegernumber of lesions detected

Lesion Data

Lesion contains all lesion-level data collected during an exam. In this file, each row represents a single finding, so there can be several rows per exam. PI-RADS scores are assigned on a per-lesion basis. These data are entered into a report by the radiologist at the time of interpretation, and extracted in a structured format by the HITI lab.

Key fields

Feature nameTypeDescription
piradsintegerPI-RADS score
lesion_contentsstringLesion description
extracted_lesion_sizefloatLesion size (cm)
extracted_lesion_volumefloatLesion volume (cc)
extracted_location_basebooleanBase level presence
extracted_location_midglandbooleanMidgland level presence
extracted_location_apexbooleanApex level presence
extracted_laterstringLaterality
extracted_PZbooleanLesion zone (peripheral)
extracted_TZbooleanLesion zone (transition)
extracted_T2stringT2-specific qualitative features
extracted_DWIstringDWI-specific qualitative features
extracted_ADCstringADC-specific qualitative features
extracted_DCEstringDCE-specific qualitative features
max_extracted_lesion_sizefloatLargest lesion dimension (cm)

Pathology Data

Pathology contains pathology and biopsy results.

Key fields

Feature nameTypeDescription
specimen_collectdatedate specimen collected
path_resstringpathology result (core biopsy)
gleason_resstringGleason score from pathology
gleason_res_surgerystringGleason from surgical pathology
gleason_res_groupintegerISUP grade group
benign_path_resbooleanBenign status indicator
diagnosis_afterstringdiagnosis after pathology integration

Staging_spread Data

Staging_spread contains information on the local tumor staging and invasion.

Treatment Data

Treatment contains surgical and therapy information.

Image Metadata

Contains image-level information and is structured as one row per file. Information includes DICOM metadata, series description, and file path.

Key fields

Feature nameDescription
empi_anonUnique anonymized patient ID. All exams for a patient will have the same empi_anon
acc_anonUnique ID per exam. All rows for an exam will have the same acc_anon (and the same empi_anon)
anon_dicom_pathAnonymized file path
study_date_anonAnonymized date of acquisition of the exam
StudyDescriptionName of the procedure
SeriesDescriptionName of the series for an exam. This can vary depending on view position.
View PositionType of view acquired (Axial, Sagittal or Coronal)

On this page