Report 62 · Patient Horizon Lung Research Plan

Transcribed source page 1 of 4

 Patient Horizon
 A lung-focused demonstration and university research plan

 Kira Labs | Evidence reviewed October 20, 2026 | Reusable discussion brief


   Purpose: Build a small, measurable research demonstration in which synthetic results update a versioned
   virtual-patient record and a transparent anatomical display. Develop organ behavior through separately
   verified numerical models and anatomical measurements.


 What can be supported today

   Area                   Recorded evidence                                   Limit


   Patient Horizon        Four synthetic request-planner tests reported in    No live hospital connection, patient-specific
                          the October 15 record.                              organ simulation or treatment prediction
                                                                              demonstrated.


   Avatar                 512 sampled weighted skin vertices moved            Geometric skin control; no proof of physiological
   engineering            through five commanded poses and returned to        lung behavior or passive tissue realism.
                          their original references.


   Body and head          A saved provisional body asset has a declared       Asset structure and preparation are distinct from
   work                   79-joint rig and anatomy references. New head       a complete, functioning anatomical body or a
                          and visitor-memory integration is being prepared.   tested Evelyn release.


   NewBrain               Original12 latest completed main result: 2/12       No net Original12 gain. Qwen comparison
                          before and 2/12 after 256 updates, with two new     remains untested. Public2 is a narrower
                          and two lost missing-information passes.            experiment.
                          Separate Public2: 1/2 to 2/2 after reload.



 A positive engineering result with a precise scope
 A separate numerical check passed 80 sampled coordinate comparisons and four direction comparisons
 across ten parameter groups; two deliberately wrong-gradient controls were rejected. It used 170 raw
 objective evaluations and made no training updates. This supports the sampled derivatives, not successful
 learning, generalization or full optimizer correctness.
 A later preparation attempt stopped before any model run because required metadata references were missing. It produced no
 new learning score; its failed record and the preserved checkpoints remain available.




KIRA LABS | Research plan | October 20, 2026                                                                                      1

Transcribed source page 2 of 4

 The first useful demonstration
 One synthetic patient, a transparent body and traceable changes

 The proposed visible sequence
 1. Open the baseline. Show available anatomy in a transparent body, identify the lungs, and display a
 synthetic patient ID, encounter, data time and record version. Use actual available meshes; label missing
 anatomy.

 2. Receive a result. Import a synthetic laboratory report or a pre-annotated scan finding. Preserve the
 source, units, status and time; show what changed. A new lab value must not invent a tumor or alter anatomy
 without supporting imaging or reviewed annotation.

 3. Replay a bedside stream. Replay timestamped synthetic heart-rate, blood-pressure, oxygen-saturation
 and respiratory-rate readings. Mark delayed, invalid or disconnected data visibly. Demonstrate latency with
 recorded send/receive/display times.

 4. Inspect breathing motion. Show the lungs from front, side and rear, with a numeric cycle trace. A
 geometric motion prototype must be labeled as such until a separately assessed physiological model drives
 it.

 5. Review research evidence. Present a small source-linked evidence summary for clinician review. Keep
 observed findings, literature and simulated outputs visibly distinct.

 Proposed checks - all remain unrun for this demonstration

   Check                            Evidence to save


   Record integrity                 Duplicate, out-of-order and corrected reports; patient/encounter mismatch rejection; previous
                                    versions remain inspectable.


   Device replay                    Timestamp and unit checks; invalid values and missing data; reconnect behavior; measured
                                    latency distribution.


   Motion and restoration           Original mesh and rig identities; commanded cycles; per-frame deformation and return error;
                                    repeated cycles without drift; multi-angle recording.


   Meaningful negative              A frozen lung, broken data link and stale feed must fail their matching checks. An intact display
   controls                         must pass the same predeclared checks.


   Clinician evidence review        Every research statement has a resolvable source/date; missing or conflicting evidence is
                                    exposed; unsupported conclusions are rejected.


 Freeze cases, tolerances, clocks and success criteria before running. Save failures, complete logs, source versions and
 cleanup evidence. A passing animation test is not a clinical or human-realism validation.




KIRA LABS | Research plan | October 20, 2026                                                                                        2

Transcribed source page 3 of 4

 How the pieces should connect
 Separate observed data, numerical physiology and generated language


   Proposed flow
   Authorized input -> validated observations -> versioned patient record -> reviewed avatar findings

   Separate numerical model -> time-stamped simulation outputs -> organ motion and graphs

   Source retrieval -> cited evidence -> language summary -> clinician review


 A physiological model should drive organ behavior
 The target is coordinated lung, diaphragm and chest motion with appropriate anatomical constraints. A
 breathing mesh cannot establish airflow, pressure, compliance, gas exchange or circulation by itself. Pulse is
 a candidate numerical physiology engine; its respiratory documentation describes driver, circuit and
 mechanics components [1]. Integration, local fit, licensing, parameter provenance and validation must be
 checked before claiming a working connection.

 For a future lung-brain study, define explicit physiological variables and direction of influence, including gas
 transport, circulation and respiratory-control signals within the selected model. Compare numerical outputs
 against documented reference scenarios. A language model is not a physiological brain, and exchanging text
 with it does not recreate human nerves or respiratory control.

 Qwen first; a qualified NewBrain interface later
 Use the existing local Qwen backend initially for source-grounded explanation and dialogue, subject to its
 own runtime and output checks. Keep the patient record, simulation engine and data interface independent
 of the language backend. Replace that backend with NewBrain only after separate capability tests; changing
 the language model must not silently change the physics or patient observations.

 An evolving record, with traceability
 Represent findings with patient and encounter identifiers, source, clinical time, issue time, units, quality and
 review status. FHIR Observation and DiagnosticReport provide relevant data structures [2,3]. Imaging
 integration additionally needs reviewed mappings and original-image references. Real device and
 medical-record interfaces require separate integration work and institutional authorization.

 Research assistance for the doctor
 Use published literature, trial records and approved local sources to retrieve evidence. Show why an item
 may be relevant, its limitations, and what still needs clinician confirmation. Follow-up investigations and
 treatment discussions are proposals for review, never automatically ordered actions. The FDA emphasizes
 the ability of healthcare professionals to independently review the basis of relevant decision-support
 recommendations [4]; the intended function still requires its own regulatory assessment.
 Current status: these interfaces and this organ demonstration are a development plan. There is no validated patient-specific
 response, diagnosis or treatment prediction.




KIRA LABS | Research plan | October 20, 2026                                                                                    3

Transcribed source page 4 of 4

 What collaboration can deliver
 A bounded research milestone with reusable evidence

 People and facilities
 Computer science and machine learning: diagnose acquisition, response formatting, retention and
 evaluation controls; run independent reproducible comparisons.
 Biomedical engineering, anatomy and physiology: select an appropriate lung model, map motion to
 anatomy and define reference scenarios.
 Clinical informatics and clinicians: review intended uses, data quality, provenance and evidence
 presentation.
 Students and research assistants: run frozen test cases, record results and verify repeatability under
 supervision.

 Computing and access requested
 Access to a dedicated research workstation with more RAM and GPU memory, a remotely accessible
 university server, storage for versioned assets and complete evidence, and scheduled laboratory time. A
 capable laptop and VR equipment could support portable demonstrations and anatomical review. Final
 specifications should follow measured workloads and the institutional security and access policies.

 The present computer has about 31.4 GiB of physical RAM. A recent start check had about 9.4 GiB
 available; the project retains an 8 GiB startup guard and 3 GiB reserve. More capacity would reduce
 resource-related interruptions and support larger assets and reproducible runs. It does not resolve a
 defective learning objective or guarantee a better score.

 Acceptance packet for the first demonstration
 Deliver a runnable synthetic case; exact source/asset and model versions; frozen test cases and tolerances;
 numeric before/after traces; failures and negative controls; restart and cleanup evidence; and a short video
 plus still images from several angles. Report three separate verdicts: software works as specified, numerical
 model agrees with its chosen references, and anatomy/motion agrees with the selected anatomical
 evidence. Leave untested claims marked pending.

 Public technical references
 1. Pulse - Respiratory Methodology
 2. HL7 FHIR R5 - Observation
 3. HL7 FHIR R5 - DiagnosticReport
 4. FDA - Clinical Decision Support Software
 5. PubMed - Data access
 6. ClinicalTrials.gov - Data API
 Project evidence: the preserved Original12 and Public2 reports; sampled-gradient audit; October 15 synthetic planner record;
 and 512-vertex movement/restoration record. Public repository research-status notes describe their scopes. No new clinical
 claim is made by this plan.




KIRA LABS | Research plan | October 20, 2026                                                                                    4