Digital Diagnostics

LumineticsCore (cleared as IDx-DR) analyzes retinal images and issues a diagnostic result for more than mild diabetic retinopathy with no physician interpretation of the image; the 510(k) summary (K213037) describes an autonomous, AI based system whose client software sends images for analysis and returns one of three results. Remove the algorithm and nothing remains but a camera and a results portal. The company also lists SphereDx, a diagnostic results platform, and states that DermSpot is not currently authorized or available for sale.

Autonomous by design, with what the system may and may not do published. LumineticsCore returns a diagnostic result at the point of care without a clinician reading the image. The boundary is written into the clearance (K213037): adults diagnosed with diabetes who have not previously been diagnosed with diabetic retinopathy, on the Topcon NW400 camera, used by healthcare providers.

The outputs are three and each routes the patient: more than mild diabetic retinopathy detected (refer to an eye care professional), not detected (rescreen in 12 months), or insufficient image quality; image quality is checked during acquisition with visual feedback to the operator. The autonomy is therefore bounded by a narrow indication, with defined behavior outside it and a defined referral action. The home page states the company sells only to healthcare providers.

The system is inspectable on the regulatory and scientific record rather than in marketing. The De Novo authorization (DEN180001) set special controls that bind every later device in its category; the 2022 clearance summary (K213037) describes an autonomous AI based system that analyzes specific disease features, client software with image quality checking and visual feedback, cloud based analysis on dedicated servers, three defined outputs and a predicate lineage (v2.0 to v2.3).

The pivotal trial was preregistered, and the founder has published on the system's design and ethical framework. Model architecture, training data composition, and how model updates are validated and deployed between clearances are not published.

The diagnostic model is the company's own cleared device, and the hosting behind it is unnamed. LumineticsCore's algorithm is authorized in the company's name (DEN180001, K203629, K213037), so it cannot be swapped without a regulatory consequence. The chain beyond the model is less clear.

K213037 describes analysis on dedicated servers in a secure datacenter without naming the operator; no subprocessor list is published; the cleared configuration depends on a named camera maker (Topcon); and the privacy notice says patient data is used to improve the algorithm, so the path from clinic to training set passes through infrastructure a buyer cannot see.

A prospective, preregistered pivotal trial in the setting of use, peer reviewed, with independent literature beside it. The pivotal study enrolled adults with diabetes at primary care sites, with endpoints fixed in advance; the 2022 clearance summary (K213037) reports its data for the current version: 850 participants, 809 diagnosable, sensitivity 87.69 percent (82.24 to 91.95) and specificity 90.07 percent (87.42 to 92.32) against a retina specialist reference, 95.18 percent diagnosable.

The SEE study (Diabetes Care, 2021) extended safety, efficacy and equity testing to a youth population outside the authorized indication. A review in Ophthalmology Science assessed LumineticsCore alongside EyeArt and AEYE Diagnostic Screening and gathered implementation experience from academic health systems, which places the product in a third party comparison. The founder is an author on much of the pivotal and framework literature, disclosed in those publications.

The company states plainly that patient data trains its algorithm, and says little else about retention. The Notice of Privacy Practices, issued by Digital Diagnostics for LumineticsCore exams, says: "We use PHI about you to improve our artificial intelligence algorithm and for FDA required monitoring of our algorithm."

That is a categorical statement on training use a patient and a buyer can hold the company to, and it settles the training question in the direction of use: retinal photographs are biometric data, captured in a primary care office from patients attending a screening test. The 2022 clearance summary (K213037) shows the path the data takes: client software sends images to analysis on dedicated servers hosted in a secure datacenter, so images leave the site.

No retention period, de identification standard or opt out is published. Safety engineering, meaning guardrails and how a safety event is handled, is not published beyond the device's image quality check.

The company states a HIPAA role for its diagnostic service and publishes a notice behind it; the provider side is not addressed. Digital Diagnostics issues a Notice of Privacy Practices for LumineticsCore exams, which covers uses and disclosures for treatment, billing and operations, public health and research, sets out patients' rights of access, correction, restriction and accounting, and gives a privacy contact; the site also describes LumineticsCore as operating as an Independent Diagnostic Testing Facility.

Issuing a notice of privacy practices is how a covered entity states its status, and the notice reaches the service a patient receives. The notice also says the company uses PHI to improve its AI algorithm and for FDA required monitoring. What is not stated is whether the company signs business associate agreements with the providers whose patients it screens, or how the covered entity role and any business associate role divide.

Nothing independent is published on security. The site links a privacy policy, terms, a notice of privacy practices, an accessibility statement and a patents page, and no security page, trust center, certification or vulnerability disclosure policy. The 2022 clearance summary (K213037) says the submission addressed data confidentiality, integrity, availability, denial of service and malware, which is the manufacturer's own premarket cybersecurity work rather than an outside check. That matters more because every exam sends a retinal image, which is biometric data, to a hosted datacenter.

The authorization created the category, and each version since is identifiable. FDA records list IDx-DR under De Novo DEN180001 (11 April 2018, filed as IDx, LLC, 21 CFR 886.1100, product code PIB), the first autonomous AI diagnostic system authorized by the FDA, granted after Breakthrough Device review, followed by 510(k)s K203629 (June 2021, filed as Digital Diagnostics) and K213037 (IDx-DR v2.3, 17 June 2022, predicate K203629).

K213037's indication is adults diagnosed with diabetes who have not previously been diagnosed with diabetic retinopathy, on the Topcon NW400 camera, with three outputs (more than mild diabetic retinopathy detected, not detected, or insufficient image quality). Other devices have since cleared against it as predicate. The home page states that DermSpot is not currently authorized or available for sale. De Novo authorization is a more demanding route than a 510(k) resting on an existing predicate.

The pivotal trial was tested across groups, and its results by group are published as conclusions rather than figures. The FDA's De Novo decision summary for the 2018 authorization (DEN180001) reports a pivotal study of 900 participants at 10 primary care sites against a reading center reference, and states that there were no significant effects of age, sex, race, ethnicity, HbA1c, lens status or site on sensitivity, and none of sex, ethnicity, race, HbA1c, lens status or site on specificity, which was higher in participants over 65. The per group sensitivities and specificities are not given, so a buyer cannot see how large any difference is.

Equity is part of the company's evidence framework: the pivotal work was framed around safety, efficacy and equity, the SEE study in Diabetes Care (2021) carried that framing into a youth population outside the authorized indication, and the founder has co authored work on bias in health AI. The 2022 clearance summary (K213037) reports no subgroup analysis. Retinal pigmentation, a plausible source of variation in retinal image analysis, is not among the groups reported.

A published error rate with its method and denominator is on the record. The 510(k) summary for IDx-DR v2.3 (K213037, June 2022) is Digital Diagnostics' own filing: a retrospective analysis of the pivotal study data, 850 participants with 809 diagnosable at final submission, against a retina specialist clinical reference, reporting sensitivity of 87.69 percent (95 percent CI 82.24 to 91.95) and specificity of 90.07 percent (87.42 to 92.32), with 95.18 percent of participants diagnosable.

It is a result on study data for the cleared indication (adults with diabetes and no prior diabetic retinopathy diagnosis, on the Topcon NW400 camera), not performance in use. The company's founder has long argued that the creator of an autonomous AI should assume liability for its diagnoses; no warranty, indemnity or liability term to that effect is published on the company's pages.

Imaging standards support is on the regulatory record, and record integration is not verifiable. The regulatory record for the current versions describes DICOM image submission, and results reach the provider through the company's systems, including SphereDx, which it lists as a diagnostic results platform. Epic appears on the home page only in a customer testimonial, not as a stated integration, and no EHR, interface standard or marketplace listing is named by the company.

The camera dependency matters: the cleared configuration names the Topcon NW400 fundus camera, so an autonomous diagnostic system carries a hardware purchase and relationship that a software discussion will not surface.

Images are captured at the point of care and analyzed in a hosted datacenter, whose location is not stated. The 2022 clearance summary (K213037) describes client software that transfers images to cloud based analysis running on dedicated servers hosted in the same secure datacenter, and a result returned to the point of care. The capture is local; the analysis is not, so a retinal image leaves the clinic for every exam. No hosting provider, region, tenancy or residency statement is published, and the home page declines to disclose which facilities house the company's products.

The payer's side is public, and the vendor's price is not. The home page states that LumineticsCore is reimbursable using CPT code 92229, for which Medicare finalized a national payment in its calendar year 2022 rules, the first national payment for an FDA authorized autonomous AI service; and the founder published a peer reviewed framework in npj Digital Medicine setting out how value and cost for the service were analytically determined.

Those describe what a payer pays the provider, not what the provider pays Digital Diagnostics. The company's own charge, its unit (per exam or per site) and any camera or platform fees are not published, and the home page states it sells only to healthcare providers.

Deliberately narrow, and validated where it is meant to operate. The authorized indication covers one finding, more than mild diabetic retinopathy including macular edema, in adults with diabetes and no prior diabetic retinopathy diagnosis (K213037), on a named camera, and the setting is primary care and other points of care where patients with diabetes are already seen, which is the strategic point, since diabetic eye screening often requires a separate visit to an eye specialist.

The pivotal trial ran prospectively at primary care sites, so the setting of use is the setting validated. Pediatric evidence exists (the SEE study) and is outside the authorized indication. DermSpot, the company's skin product, is stated not to be authorized or available.