Skip to content

About

application of interconnected programs for each body system... sans substrate!

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

BjorkEd Body Lab

An educational body-system simulation with private, session-only health learning workspaces. This is not a clinically validated digital twin, diagnostic medical device, radiology workstation, or replacement for a clinician. Simulated values are not measured health. Citations document educational sources, not clinical validation of this software.

Run

Python 3.10+ and a modern browser are sufficient to run the committed app; no package installation or AI-provider credentials are required. The 3D viewer requires WebGL2. If it is unavailable, the app explicitly reports that limitation and the accessible organ lessons remain usable.

python3 server.py --port 4790
# Open http://127.0.0.1:4790/

The server binds to loopback by default. Serve the same allowlisted static assets over trusted HTTPS for a phone; browser file access and home-screen installation depend on browser support. The web manifest supports a standalone launch experience, not native HealthKit/Health Connect synchronization or an App Store/Play release. No service worker caches personal data. The former incomplete GKE deployment template was replaced with executable checks; no cloud deployment is claimed.

To modify the anatomical renderer, use Node.js 22+, run npm ci, then npm run build. Commit the generated js/body-model.bundle.js alongside its sources. The bundle includes pinned three.js and its MIT license; no runtime CDN requests are made. CI rebuilds it before running the behavior suites.

Workspaces

Simulation and anatomy

  • Fourteen existing system agents plus distinct pancreas, liver and kidney agents, with visible cards and pause/resume controls.
  • 3D body & anatomy uses licensed BodyParts3D reference meshes, replacing the former ellipsoid mannequin and generated skin. The asset contains 1,179 source anatomical components: skin, whole/segmented muscles and tendons, bones, brain, visceral organs and vascular/airway structures. Detailed facial contours, hands, feet, muscle boundaries and organ surfaces come from the dataset, not invented primitive geometry.
  • Switch between musculature, body surface, internal organs, skeleton, combined layers and Selected organ only. Organ buttons isolate and frame the real heart, lungs, liver, pancreas, kidneys or small intestine and open the corresponding lesson/quiz. Clicking an organ in the contextual view selects it without losing that view. Drag or use arrow keys to rotate; wheel/pinch or +/− to zoom; camera buttons and Home/Reset restore framing.
  • The default three-quarter muscular view includes the underlying skeleton. Rough tissue materials, studio environment lighting and ground shadows make contours legible. Rendering is event-driven, not an idle animation loop. The local ~37 MB GLB loads only when the anatomy workspace is opened; it makes no runtime dataset/CDN requests and is not constructed from personal measurements or uploaded scans.
  • This is an adult male reference, including external anatomy in the surface layer, not a personalized or population-representative body. Colors, lighting and selection highlights are illustrative. Display scale and simplifier error are not clinical measurements. The selected/optimized meshes omit structures and individual variation; there is no physiological motion or clinical validation.
  • Pancreas owns insulin/glucagon signals, liver owns its glycogen store and a separate urea production pool, kidneys clear that urea pool. Endocrine retains peripheral glucose uptake; Excretory displays renal state rather than removing it twice.
  • Equations, ownership and omissions are documented inside 3D body & anatomy. These are hand-chosen toy dynamics, not research-fitted human physiology. Arbitrary-unit pools and a 1:1 illustrative glycogen/glucose mapping are not clinical concentrations or physical mass conversion.
  • Scenario condition switches are hypothetical configuration, not diagnoses inferred about the user. The homeostasis index is a toy score, not a health assessment.

Anatomical asset provenance and rebuilding

BodyParts3D, © The Database Center for Life Science licensed under CC Attribution 4.0 International. See the official dataset, current publisher license and CC BY 4.0. The publisher changed the license in February 2025; older OBJ file comments retain the previous CC-BY-SA-2.1-JP text.

The geometry uses release 4.0 plus the five original release 3.0 pulmonary lobe surfaces: the 4.0 PART-OF lung hierarchy otherwise supplies airways/vessels without the enclosing lobes. Both retain their source coordinate relationship; no invented lung shell is substituted. assets/anatomy/credits.json records source URLs, input/per-lobe hashes, component identifiers, modifications and the final asset hash.

Source geometry is coordinate-normalized, colored, batched and simplified offline with locked mesh boundaries. About 3.92 million original triangles become 1.20 million rendered triangles in 22 batches. This reduces transfer/GPU cost without substituting the former procedural body; the simplifier's numerical error is not an anatomical accuracy guarantee.

To regenerate the committed asset, download isa_BP3D_4.0_obj_99.zip, partof_element_parts.txt and isa_element_parts.txt from the official current archive, and BodyParts3D_3.0_obj_99.zip from the official 3.0 archive. After npm ci:

python3 build-anatomy.py \
  --archive /path/to/isa_BP3D_4.0_obj_99.zip \
  --partof /path/to/partof_element_parts.txt \
  --isa /path/to/isa_element_parts.txt \
  --lung-archive /path/to/BodyParts3D_3.0_obj_99.zip
node optimize-anatomy.mjs
npm run build

The lung input may instead be a ZIP containing the five unmodified original entries FMA7333, FMA7337, FMA7370, FMA7371 and FMA7383. Optimization uses pinned meshoptimizer only during development; no WASM decoder or relaxed browser CSP is needed. Commit the GLB, provenance JSON and renderer bundle. The old procedural body/surface generator has been removed.

Wearables and personal health

Consent → local validation → preview → explicit apply. The applied resting pulse becomes the educational baseline; step count never determines a fitness label. Other supported measurements are available to the assistant only with a separate opt-in.

Supported imports:

  • Apple Health XML: select the unzipped export.xml. Resting heart rate, steps, supported asleep categories and oxygen saturation. HealthKit percent samples use fractions (0.98 means 98%).
  • Defined CSV, with exact header metric,value,unit,start,end. Metrics are restingHeartRate, averageSteps, sleepHours, oxygenSaturation; units and timezone-aware date rules are documented in the UI. CSV percentage values use 0–100.
  • FHIR Observation or Bundle JSON with the documented recognized codes, units, timestamps and one subject. Unsupported codes are not guessed into known metrics.
  • Manual measurements use the same validation and preview flow. Unknown values remain unknown, not zero.

Limits: 12 MiB/file and 100,000 records. Latest resting pulse/oxygen samples are used, not a clinically measured long-term baseline. Steps/sleep are averaged over observed UTC days across the file; missing days are not zero. Exact duplicates are removed; conflicting/overlapping samples, mixed FHIR subjects, invalid units and impossible values are rejected. The UI reports provenance and stale-data limitations. Native watch/account pairing requires future platform SDK, permission and device validation by the native departments.

Personalized health assistant

A transparent offline, fixed-rule English-language conversational assistant, not a hidden cloud LLM. It covers heart/breathing, fatigue/sleep, headache, digestive symptoms, exercise/wearables and stress/mood. It gathers duration, severity, progression and relevant context; offers possible explanations and low-risk education; and routes uncertainty and concerning symptoms to professional care.

Medical emergencies and self-harm concerns take precedence. Normal wearable readings never override concerning symptoms. The assistant cannot rule out emergencies, make confirmed diagnoses, prescribe or change medication/oxygen, interpret scans, monitor someone, or summon help. It is not comprehensive screening; phrase recognition and negation are limited. Pediatric, pregnancy and serious-condition contexts need clinician review. The interface exposes sources and limitations. A diagnostic product requires qualified clinical validation and appropriate regulatory, security and privacy review; none is implied by this app.

Medical images

Files are selected into browser memory, not uploaded to a server. View/delete PNG and JPEG images, or supported DICOM Part 10, with zoom and contrast controls.

DICOM support: single-frame, 8/16-bit MONOCHROME1/2, Implicit/Explicit VR Little Endian; signed/stored bits, rescale and LINEAR windowing. Maximum 32 MiB/file, 4096 pixels/side and 8,388,608 pixels. Unsupported compressed/encapsulated/deflated/Big Endian, multiframe, color/palette, floating point, LUT/nonlinear presentation and raw non-Part-10 studies are rejected explicitly. Animated PNG is unsupported.

No disease detection, radiology interpretation, orientation correction or calibrated clinical measurement. Patient tags/filenames are not listed, but identifiers burned into pixels can still be visible. Deletion releases/zeros retained buffers and canvases; JavaScript cannot promise secure erasure of browser/OS copies or original files.

Mental wellbeing

Optional consented mood/stress/sleep check-ins and a session journal, individual deletion, self-paced sensory grounding and a stoppable gentle breathing exercise. Exercises stop when leaving the workspace or hiding the tab. Check-ins are unscored self-reports, not diagnoses or crisis monitoring. Crisis resources include local emergency guidance, US 988, Samaritans and country-based support. Breathing exercises are not for acute breathing difficulty.

Development departments

Thirteen real recurring Hermes jobs were configured on the owner's VPS: Organ Systems Research, Wearable Integration, Sales, iOS Engineering, Android Engineering, Web Engineering, Art & Visualization, Education, Finance, Research, Mental Health, Clinical Safety & Chat, and Medical Imaging.

data/departments.json is a timestamped deployment snapshot, not live worker status. The Departments workspace shows actual configured job IDs, missions and Monday/Thursday schedules (scheduler timezone, UTC−05:00 when configured). Reports live under /root/reports/body-lab/<department>/ on that private VPS, not in this public app. The Business Village also observes the jobs in its Body Lab room. These agents research and develop the product with synthetic data; they have no access or authority to process user health records. Native teams are deployed, but no native build, device integration, signing or store publication is claimed.

Privacy and security

  • Consent is off by default; imported summaries, chat, image data and wellbeing entries stay in the current page's memory. No cloud AI calls, analytics, IndexedDB or local/session storage for health data.
  • Clear all personal data, consent withdrawal, local clear and page exit erase relevant app-held state. Reloading loses the session. External source links only open when deliberately selected; their sites have separate privacy policies.
  • The local server serves allowlisted assets, denies private/repository files and all POST/PUT/PATCH/DELETE health routes, uses no-store, and restricts network connections through CSP. It is not a multi-user healthcare backend.
  • Browser extensions, keyboard services, OS memory, screenshots and shared devices remain outside the app's guarantees. Do not use someone else's medical records without authorization. No HIPAA/GDPR compliance certification is claimed.
  • The legacy health-aggregate upload, randomly mutated “research” parameters, their demo data and the obsolete standalone prototype entrypoint were removed. Reviewed source/code updates replace synthetic research claims.

Verification

Node.js 22+ runs the behavior tests and renderer build; Python runs the HTTP boundary checks:

npm test
python3 -m unittest discover -s tests -p 'test_*.py'

The suites cover importer boundaries, organ transfer invariants, symptom escalation/uncertainty, DICOM parsing/rendering and HTTP privacy boundaries. Synthetic fixtures are used; no real patient data is required.

Integrated Chromium checks exercised Apple XML import/preview/apply, baseline personalization, educational quiz feedback, organ response to food, emergency precedence despite normal wearable values, DICOM pixel rendering, wellbeing check-ins, stopped exercises, cross-workspace consent/data clearing, department cards and 390px layouts. This is engineering verification, not clinical validation or native iOS/Android device testing.

The reference-mesh viewer was checked in Chromium for all six isolated lessons, contextual lung raycast selection, surface/muscular/skeletal/combined layers, camera presets/zoom, keyboard rotation/reset, 390px layout and absence of external requests. Opening the simulation alone does not fetch the GLB. The HTTP regression suite verifies that the exact public asset is served without exposing arbitrary medical-model files. Visual realism does not establish clinical validation.

About

application of interconnected programs for each body system... sans substrate!

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages