fundus-camera-watchdog 0.3.0

File watchdog for fundus camera

Fundus Camera Watchdog

camera_watchdog.py monitors a folder on the fundus camera workstation and uploads files to a CLINICEDC project using the edc-retinopathy API.

It is designed to run on the camera’s workstation. When the camera finishes an examination and writes files to disk, the watchdog detects them, resolves the subject against the CLINICEDC server, uploads each file, and moves the completed folder to an archive folder.

Prerequisites

1. A running CLINICEDC server with edc-retinopathy installed and an API token created.

2. The camera software must write its output into the watched folder using the expected layout (see Folder layout below).

3. The camera’s SQLite database must be accessible from the workstation (typically a local file).

Folder layout

The camera creates one subfolder per subject, named with the subject identifier. Inside each subfolder are UUID-named files.

Combined report (default — one HTML covers both eyes):

C:\RetCamOutput\
    105-10-0989-3\
        a1b2c3d4.jpg          <- eye image (left or right)
        e5f6a7b8.jpg          <- eye image (left or right)
        c9d0e1f2.html         <- combined report (both eyes)
    105-10-0001-2\
        ...

Per-eye report (one HTML per eye):

C:\RetCamOutput\
    105-10-0989-3\
        a1b2c3d4.jpg          <- eye image (left or right)
        e5f6a7b8.jpg          <- eye image (left or right)
        c9d0e1f2.html         <- eye report (left or right)
        f3a4b5c6.html         <- eye report (left or right)
    105-10-0001-2\
        ...

Because filenames are random UUIDs, the watchdog queries the camera’s SQLite database to determine which file belongs to which eye.

Processing is triggered once a subject folder contains the expected number of files:

• Combined (default): at least 2 JPEG and 1 HTML file (3 files total).

• Per-eye: at least 2 JPEG and 2 HTML files (4 files total).

Configuration

All settings live in a single JSON file. Create camera_config.json:

{
    "watch_dir": "C:\\RetCamOutput",
    "db_path": "C:\\RetCamOutput\\camera.db",
    "api_url": "https://edc.example.com",
    "token": "YOUR_DRF_TOKEN",
    "device_id": "RET-CAM-001",
    "site_id": "40",

    "db_patient_table": "patients",
    "db_patient_subject_id": "subject_identifier",
    "db_patient_initials": "initials",
    "db_patient_sex": "sex",
    "db_patient_age": "age",

    "db_image_table": "images",
    "db_image_subject_id": "subject_identifier",
    "db_image_filename": "filename",
    "db_image_eye": "eye",

    "report_type": "combined"
}

Required keys

watch_dir

Folder the camera writes subject subfolders to.

db_path

Path to the camera’s SQLite database.

api_url

Base URL of the EDC server (e.g. https://edc.example.com).

token

DRF authentication token for the camera user.

Optional keys

device_id

Identifier for this camera (sent to the server with each session).

site_id

Study site identifier.

report_type

How the camera writes its analysis reports. combined (default) means a single HTML file covers both eyes; per_eye means one HTML per eye. This controls how many files the watchdog expects before triggering an upload (3 for combined, 4 for per_eye).

log_level

One of DEBUG, INFO (default), WARNING, ERROR.

Database column mapping

These keys tell the watchdog which tables and columns to query in the camera’s SQLite database. Adjust them to match your camera vendor’s actual schema.

Config key	Purpose	Default
db_patient_table	Table containing patient demographics	patients
db_patient_subject_id	Column for subject identifier	subject_identifier
db_patient_initials	Column for initials	initials
db_patient_sex	Column for sex (M/F)	sex
db_patient_age	Column for age in years	age
db_image_table	Table mapping files to eyes	images
db_image_subject_id	Column for subject identifier	subject_identifier
db_image_filename	Column for UUID filename	filename
db_image_eye	Column for eye laterality	eye

The watchdog normalises eye values automatically. All of the following are recognised:

• Left eye: L, LE, OS, LEFT

• Right eye: R, RE, OD, RIGHT

Usage

With a config file (recommended):

uv run camera_watchdog.py --config camera_config.json

CLI flags override any value from the config file:

uv run camera_watchdog.py --config camera_config.json --log-level DEBUG --report-type per_eye

Without a config file (all flags on the command line):

uv run camera_watchdog.py ^
    --watch-dir C:\RetCamOutput ^
    --db-path C:\RetCamOutput\camera.db ^
    --api-url https://edc.example.com ^
    --token YOUR_TOKEN ^
    --device-id RET-CAM-001 ^
    --db-patient-table Exams ^
    --db-patient-subject-id patient_code ^
    --db-image-table CapturedFiles ^
    --db-image-eye laterality

Stop with Ctrl+C.

What it does

The watchdog runs continuously and performs the following for each subject:

1. Detect — watches for new files in subject subfolders using filesystem events (watchdog), plus a periodic sweep every 60 seconds as a safety net.

2. Wait — each file is given up to 30 seconds to stabilise (stop growing) before being registered, so half-written files from the camera are not picked up prematurely.

3. Query camera DB — retrieves the subject’s demographics (initials, sex, age) and determines which JPEG is the left eye and which is the right. In per-eye mode, the HTML reports are also mapped to eyes; in combined mode the HTML is uploaded without eye mapping.

4. Resolve — POST /api/retinopathy/resolve/ validates the subject against RegisteredSubject on the CLINICEDC server and creates (or reactivates) a session.

5. Upload — sends each file to the correct API endpoint:

   • *.jpg + left eye -> POST .../left/

   • *.jpg + right eye -> POST .../right/

   • *.html (combined) -> POST .../report/

   • *.html + left eye (per_eye) -> POST .../left_report/

   • *.html + right eye (per_eye) -> POST .../right_report/

   Each upload includes a SHA-256 checksum for integrity verification.

6. Verify — GET .../status/ confirms the session received all expected files.

7. Archive — the entire subject folder is moved to <watch-dir>/processed/<subject_id>_<timestamp>/.

Error handling

• Retries — every API call is retried up to 3 times with a 5-second delay between attempts.

• Failed subjects — if any step fails (server unreachable, validation error, upload failure), the subject is marked as unprocessed and will be retried on the next 60-second sweep.

• Startup scan — on (re)start the watchdog scans all existing subject folders, so a restart after a crash picks up where it left off.

• Thread safety — file detection and upload run on separate threads with proper locking, so multiple subjects can be uploaded concurrently.

Inspecting the camera database

To discover your camera’s actual table and column names, open the SQLite database and list its schema:

sqlite3 camera.db
.tables
.schema patients
.schema images

Then update the db_* keys in your config file to match.