Data record for the article: Deep learning for diagnosis of Acute Promyelocytic Leukemia via recognition of genomically imprinted morphologic features

This
metadata record provides details of the data supporting the claims of the
related article: “Deep learning for diagnosis of Acute
Promyelocytic Leukemia via recognition of genomically imprinted morphologic
features”.

The
related study aimed to demonstrate a deep learning method to assist with the diagnosis
of Acute Promyelocytic Leukemia (APL), which is a subtype of Acute Myeloid
Leukemia (AML). Deep learning pattern recognition was applied to the peripheral
blood smear haematopathology images, which are usually taken at clinical
presentation.

Type of data:
Images of peripheral blood smear results from AML and APL patients; Metadata
table for the images

Subject of
data: Retrospectively identified APL and AML patients presenting at Johns
Hopkins Hospital

Sample size:
106 study participants

Population
characteristics: APL patients were separated into a discovery cohort presenting
prior to 1/2019 (n = 22) and a validation cohort presenting on or after 1/2019
(n = 12). AML patients were separated into a discovery cohort presenting prior
to 1/2019 (n = 60) and a validation cohort presenting on or after 1/2019 (n =
12).

Recruitment:
Patients with APL were identified via retrospective chart review from a list of
confirmed FISH t(15;17)-positive patients presenting at The Johns Hopkins
Hospital (JHH) who met the inclusion criteria (n = 34) of presentation at the
time of initial diagnosis, without history of remission, presentation prior to
treatment initiation, and availability of peripheral blood smear image.

Patients
with AML were identified via retrospective chart review from a list of patients
presenting to JHH who at initial presentation had a bone marrow biopsy showing
>20% blasts and by acquiring a query of patients who tested negative for the
t(15;17) translocation by FISH and who were then confirmed to have AML by bone
marrow biopsy and other genetic studies.

Geographic
location: United States of America

Data
access

The peripheral
smear images from 106 de-identified study participants are openly available as
part of this figshare metadata record.
All images are in jpg format.

Within the ‘blood
smear images_Patient00-105.zip’ file, the images are arranged into folders
named ‘Patient_00’ to ‘Patient_105’. Within each participant folder images are divided
into subfolders named ‘Unsigned slides’ and ‘Signed slides’. The signed folder contains the images organized by Cellavision cell type, and the unsigned folder contains the images not subdivided by Cellavision cell type calls. The ‘Unsigned
slides’ and ‘Signed slides’ folders are further subdivided into folders containing
images of the specific peripheral blood cell types.

The metadata
table for the images is openly available as part of this figshare metadata record, ‘Images_metadata_table.csv’. This metadata
table provides the ‘diagnosis’, ‘cohort’, ‘age at diagnosis’ and ‘gender’ for
each of the participant IDs.

Corresponding author(s) for this study

Eugene
Shenderov ([e-mail]).

Study approval

All
experiments were conducted in accordance with the Declaration of Helsinki and
the International Ethical Guidelines for Biomedical Research Involving Human
Subjects. The Human Research Ethics Committee, Johns Hopkins University School
of Medicine approved the study.