Point-of-Care Gastric Ultrasound in Healthy Volunteers: Cine Loops and Still Images Across Four Controlled Gastric Content States 1.0.0

File: <base>/README.md (2,968 bytes)
# Point-of-care gastric ultrasound in healthy volunteers

720 point-of-care ultrasound acquisitions of the gastric antrum from 30 fasted
healthy adult volunteers, each imaged in four investigator-controlled gastric
states and in two body positions. Full description, methods and citation are on
the PhysioNet project page.

## Layout

    subjects.csv               one row per participant (30)
    acquisitions.csv           one row per released file (720)
    acquisition_settings.txt   machine settings, which the sector crop removes from the images
    images/<subject>/          the acquisitions
    code/                      sector mask and the script that produced images/

## Naming

`<subject>_s<state>_p<position>_i<acquisition>`

    s1 fasted        s2 clear fluid      s3 solid, early     s4 solid, late
    p1 supine        p2 right lateral decubitus
    i1 still image   i2 cine loop        i3 cine loop

Filenames are lowercase. The same codes appear in uppercase in the `state`,
`position` and `acquisition` columns of `acquisitions.csv`, and each file's
SHA-256 checksum is in its `sha256` column.

Stills are 934x766 lossless PNG; cine loops are 934x766 H.264, about five
seconds at 30-32 frames per second (nominally 31; the per-file rate is in
`acquisitions.csv`).

## What the images contain

The acoustic sector and nothing else. The crop removes the acquisition date and
time, the session code, the vendor and hospital logos, the machine settings, the
depth scale, and the operator's on-screen annotation of gastric state and body
position. There is no burned-in text for a model to read instead of the anatomy,
so no masking step is needed before training.

Two consequences. The machine settings are in `acquisition_settings.txt`, with
the fields that varied given per file in `acquisitions.csv`. And there is no
depth scale in the image: the imaging depth was 166 mm throughout, which puts
the scale at about 4.33 pixels per millimetre.

## Repeated measures

Every participant contributes 24 acquisitions across the four states and two
positions, and consecutive frames within a cine loop are near-identical.
Analyses that treat frames or acquisitions as independent will be optimistic.
Group by `subject_id`.

## Labels

Gastric content was controlled by the investigator, so each file is labelled
with the protocol state that applied when it was taken: what the participant had
eaten or drunk, and when. That is the ground truth in this version. It is not a
reading of the image, and the two can diverge — a fasted stomach may hold a small
volume of normal secretion, and a modest meal may have substantially emptied by
70-90 minutes in a healthy young adult, so some S4 acquisitions will show little
solid content.

A future version will add state labels assigned independently by experienced
operators reading the images blind, which will allow expert reading to be
compared against the controlled state and against other operators.