Optimus Diagnostics
Medicine moved into the home.Interpretation did not.
We are building an explainable interpretation layer for medicine: a concept-learning engine designed to read what a camera or probe already sees, on the device already in the patient's hands, and to state the reasoning behind every output.
Development stage. Optimus Diagnostics is a pre-revenue company. No Optimus product is cleared or authorised by the FDA or any other regulator, and nothing described on this site is available for clinical sale or for diagnostic use. Product descriptions are statements of intended development, not of present capability.
The gap
Seeing without sight
Care has moved out of the hospital and into the home and the clinic. The contact is there, the camera is there and the clinician is there. The interpretation is not, and that is the gap this company exists to close.
What the point of care already has
The hardest asset to build
- Recurring, consented contact with the patient, in the home and in the clinic
- A smartphone or a connected device already in the patient's hands
- Licensed clinicians and pharmacists with a standing reason to make contact
- A dispensing and medication record of what the patient was given
- Payor relationships written on readmission, adherence and outcome
What is missing
The interpretation layer
- Confirming the right dose was taken, not only that a lid opened
- Reading the skin of a patient on chronic immunosuppression
- Catching a pressure injury at stage one rather than stage three
- Recognising the drug eruption that says a regimen is causing harm
- Interpreting a cardiac study where no cardiologist is standing by
The engine
Explainable by design
Optimus licenses a concept-learning engine that is not a neural network. It learns a clinical concept from a small number of real examples and states the properties behind every output. Four characteristics matter in medicine, and conventional deep learning struggles to deliver them at the same time.
Few-shot
Learns a concept from a handful of real examples rather than millions of labelled images.
Explainable
Names the properties behind each output, so a clinician reviews the basis rather than a score.
Edge-native
Designed to run on a single low-power processor. No cloud round trip, and no image need leave the device.
Robust
Built for poor lighting, occlusion, motion and the uncommon presentations that thin out training sets.
Where conventional models run out of data
A trained model needs volume, so it performs on common presentations in well-photographed populations and degrades on rare disease, on darker skin tones, on paediatric anatomy and on the long tail of generic drug appearances. Those are the cases where the clinical need is greatest.
Why the reasoning matters more than the score
A pharmacist, a payor and a regulator each have to stand behind a decision. An output that cannot say why it was produced is difficult to review, to adjudicate and to submit, whatever its accuracy. Stated reasoning is a condition of operating here, not a feature.
The underlying engine is licensed from a third-party licensor and is not owned by Optimus Diagnostics. Characteristics described above reflect the licensor's technology as applied to the fields Optimus licenses.
The architecture
No data centres. No GPUs.
Conventional medical artificial intelligence is a deep neural network. It performs billions of matrix operations, which standard processors cannot do quickly, so it needs graphics hardware or a cloud cluster. A concept-learning engine is not a neural network, and that single difference is what allows it to run inside the device.
Conventional deep learning stack
The licensed engine
Concepts, boundaries and rules with graded confidence values, in place of billions of statistical weights.
No connection required
A portable probe or field monitor can work in a rural clinic or a basement with no signal.
No data leaves
Patient information can stay on the device, which removes a class of breach exposure and simplifies compliance.
No round trip
Results land locally rather than waiting on a server across the country.
No new hardware
Designed for the standard processors manufacturers already ship, with no graphics hardware to buy or power.
The platform
One engine, three clinical cores
The expensive work is the perception core: the image pipeline, the concept library, the explainability layer and the clinical data infrastructure. Built and validated once, it is then pointed at separate clinical questions, each of which carries its own validation and its own regulatory route.
DoseScope
Medication verification at the moment the dose is taken, inside a pharmacist-supervised workflow. The engine identifies what is present and checks it against the dispensed record the partner already holds.
- Identify the tablet or capsule at the dose
- Confirm ingestion under pharmacist supervision
- Reconcile the dose against what was dispensed
DermScope
A skin surface perception core intended to support four clinical questions from one validated layer, beginning with wound assessment, which carries the shortest regulated path.
- Wound pressure injury and diabetic foot ulcer assessment
- Safety cutaneous adverse drug reaction surveillance
- Detect skin lesion triage support
- Monitor longitudinal change over time
CardioScope
Cardiac imaging and rhythm interpretation intended for handheld and point-of-care hardware, where the study is captured far from the reader who would normally interpret it.
- Echo function estimates from a handheld probe
- Rhythm atrial fibrillation and conduction
- Flow valve assessment on Doppler
Product names and applications describe development intent. No product listed above is cleared or authorised for clinical use, and each is subject to regulatory review before any commercial availability. Sequencing and timing are estimates and are not commitments.
Deployment
Software that arrives with its distribution
The engine is designed to run on assets a partner has already built and paid for. That is the difference between clinical software that has to buy its way to the patient and software that arrives inside a channel which already reaches them.
| What the partner already has | What it makes possible | Why it is hard to assemble later |
|---|---|---|
| A connected device in the patient's home | Medication verification and skin surface capture | Hardware placement is slow and already paid for |
| A pharmacy or dispensing operation | Verification at the point of dispense | Every dose already leaves a record that can be compared |
| Licensed clinicians inside the workflow | All three cores | Every output needs a human with the authority to act |
| A specialty or complex-therapy population | Cutaneous surveillance on a high-risk cohort | Assembled patient by patient over years |
| Recurring, consented in-home visit cadence | Longitudinal change tracking | Longitudinal imaging needs physical access over time |
| An existing quality and regulatory system | Every submission in the plan | Quality and post-market surveillance already stood up |
One new element
The engine is inserted into a sequence that already runs, and changes nothing on either side of it.
No new data collection
The comparison uses the dispensed record the partner already holds. Nothing new is collected from the patient.
No new workflow
A discrepancy enters the clinical queue that exists today, in the same shape as every other alert a pharmacist works.
Clinical foundation
Built with the clinicians who use it
The engine's founding medical application was developed in wound healing research, and each product is designed around a step a clinician already takes. That is the difference between a tool that is adopted and one that is bypassed.
Clinical advisor
John W. Harmon, MD
Wound healing and surgical research, Johns Hopkins Medicine. The engine's founding medical application was developed with him, and he is the clinical anchor for the DermScope validation programme.
Clinical advisor
S. M. Hosseini, MD
Clinical workflow design, Johns Hopkins Medicine. Each product is built around a step a clinician already takes, rather than adding one.
Validation before contract
Concordance work is designed to run inside a partner's own operation, against a verification step the partner already performs.
Hard cases named in advance
Reformulated generics, split tablets, worn imprints and look-alike pairs are specified before a study opens rather than excluded afterwards.
Silence is the default
A reading that reconciles writes a record, not an alert. Alert fatigue is the common failure mode in this category.
Institutional affiliations are shown to identify an individual's professional role. They do not indicate review, sponsorship or endorsement of Optimus Diagnostics by Johns Hopkins Medicine or any other institution.
Regulatory position
Where the line sits
We would rather state this plainly than have a partner discover it in diligence. The first deployment is designed to reach clinical use without a new clearance because of how it is configured, and every product beyond it carries a submission.
On this side of the line
A verification aid inside a pharmacist-supervised workflow. The engine presents an identification and the reasoning behind it to a licensed pharmacist who is already reviewing that patient's therapy. The pharmacist remains the decision maker and can independently review the basis rather than rely on the output.
What would cross it
Autonomy, or a clinical claim. The position changes when an output drives an action with no clinician in the loop, when the software makes a claim about a disease or a condition, or when the clinician cannot reasonably review the basis for the result. Each of those is a regulated device function and carries a submission.
Statement of status
Optimus Diagnostics holds no FDA clearance, authorisation or approval. No product described on this site is currently available for sale, for clinical use or for diagnostic use in the United States or in any other jurisdiction.
Descriptions of DermScope and CardioScope, and of applications beyond the supervised verification configuration, describe intended development. Each requires a regulatory submission and a determination by the relevant authority before it may be marketed. Indicative sequencing reflects management's current expectations and may change.
Nothing on this site is medical advice, and nothing here is intended to support a clinical decision. Patients should speak with a qualified clinician.
The company
Who is building this
Optimus Diagnostics licenses a horizontal perception platform into defined medical fields of use and builds the clinical, regulatory and commercial layer around it.
Co-founder
Dr. Rob Sobhani
Business growth, partnerships and route to market. Founder and chief executive of Sparo, and the commercial engine behind the route to the first partners.
Co-founder
Gary LaDrido
Corporate development and capital markets. Formerly a Vice President on the Morgan Stanley Clean Energy team, with two decades of impact investment banking behind him.
Contact
Partner and platform enquiries
We are speaking with home-based care platforms, specialty pharmacy networks, connected device manufacturers and point-of-care hardware companies whose installed base already reaches the patient.
Gary LaDrido, Co-founder
What is useful in a first conversation
- The population your platform already reaches, and how contact recurs
- The hardware already placed with that population
- Whether licensed clinicians sit inside your workflow today
- The quality and regulatory system already in place