Diagnostics & Imaging

Outlook

Scope note: “Diagnostics & Imaging” means AI software for detection, triage, quantification, reporting, workflow orchestration, or population screening on diagnostic images — excludes underlying CT/MRI/X-ray/ultrasound/pathology-scanner hardware unless AI is sold as part of the imaging solution. Global/US estimates measure AI in medical imaging specifically; the most defensible India estimate measures the broader “AI in medical diagnostics” (also includes in-vitro diagnostics, predictive diagnostics, non-imaging applications) — not directly comparable.

Data quality flag: a second India estimate (MarketsandMarkets) appears internally inconsistent on its own public page, showing both $791.2M–$4.529B and $21.63M–$110.61M for 2025–2030 — should not be used for valuation without obtaining the underlying report.

Interpretation: investable software market growing ~31–35% globally, ~33% in the US under medical-imaging-specific definitions. India is directionally high-growth but published estimates vary by an order of magnitude depending on pathology/IVD/predictive-analytics/hardware/services inclusion — underwrite India bottom-up via addressable scan volumes, price per study, hospital budgets, government screening tenders, and export revenue rather than a single syndicated estimate. For context, India’s overall medical-imaging equipment and services market was estimated at $1.522B in 2025 — much larger than the AI-only pools above, showing substantial headroom but also that AI vendors capture only a limited share of workflow economics today.

AI Maturity

Verdict: proven at scale in radiology, but not uniformly mature across applications.

Radiology is the most mature clinical AI segment. In 2025, an Innolitics analysis identified 295 US FDA 510(k) clearances for AI/ML devices, of which 71.5% were radiology devices, 62% Software as a Medical Device, 63% diagnostic (methodology used AI to identify devices and may have missed some — treat as directional, not an official FDA denominator). FDA’s own AI-enabled-device list is explicitly non-comprehensive and periodically updated — any cumulative clearance count is a dated snapshot, not a definitive total.

Production deployment evidence:

• Aidoc: 45M+ patients/year across 150+ health systems; enterprise rollouts at Advocate Health, Sutter Health; deployments at Mount Sinai, Yale New Haven, University of Miami, Temple Health
• Rad AI: thousands of radiologists at organizations representing ~50% of US medical-imaging volume; four investing health systems collectively operate 100+ hospitals and were implementing system-wide adoption (company-reported, not independently audited)
• Qure.ai: 5,500+ sites, 105+ countries, 45M+ lives affected (company-profile source, should be validated in diligence)
• DeepTek: 2,000+ global sites, 3M+ lives touched; selected for deployment across Singapore’s public-sector radiology system

India-specific regulatory maturity is harder to quantify — no authoritative public CDSCO database provides a clean count of AI imaging/SaMD clearances. This absence should not be read as zero approvals; it means the evidence base is less transparent than the FDA list and requires product-by-product verification through CDSCO licenses and SUGAM/medical-device portals.

Maturity by use case:

• Most mature: acute CT triage, chest X-ray screening, breast imaging support, worklist prioritization, reporting assistance, incidental-finding follow-up
• Scaling: enterprise AI orchestration platforms deploying multiple algorithms through common PACS/RIS integration
• Earlier-stage: unconstrained foundation models, autonomous diagnostic interpretation across multiple organs, adaptive continuously-updating models, generative systems directly influencing treatment without a clinician in the loop

No longer a pilot-only category — but many point algorithms remain commercially subscale despite obtaining clearance.

Key Players

Aidoc — enterprise clinical-AI platform detecting suspected acute findings across radiology, cardiology, neurovascular, vascular services; aiOS positioned as infrastructure for deploying/governing both Aidoc and third-party algorithms. 45M+ patients/year, 150+ health systems. June 2026: $150M financing incl. $40M revolving credit, led by General Catalyst and Square Peg with NVentures and four US health systems; total funding reported at $370M. 2022: TCV and Alpha Intelligence Capital co-led $110M Series D (funding to $250M at that point). Private, VC/growth-backed; no investor stake percentages disclosed.

Rad AI — focuses on radiologist reporting/follow-up workflow: Rad AI Reporting (draft report content), Rad AI Impressions (impression sections from dictated findings), Rad AI Continuity (incidental-finding follow-up tracking) — differs from pixel-level-detection-focused vendors. Jan 2025: $60M Series C led by Transformation Capital at a reported $525M valuation (total investment $140M+); May 2025: Advocate Health, Memorial Hermann, Corewell Health, Atlantic Health System added $8M, taking the round to $68M. Thousands of radiologists at organizations representing ~50% of US imaging volume. Private, growth-equity/VC-backed (Transformation Capital, Khosla Ventures, World Innovation Lab, UP2398, Kickstart Fund, OCV Partners, participating health systems); stakes undisclosed.

Qure.ai — imaging algorithms for chest X-ray, head CT, chest CT, musculoskeletal imaging; identifies/quantifies TB, lung nodules, stroke, traumatic brain injury, fractures; combines acute-care and high-volume public-health screening use cases, particularly relevant to India and specialist-shortage markets. 5,500+ sites, 105+ countries, 45M+ lives. 2024 Series D: $65M from 360 ONE Asset, Lightspeed, HealthQuad, Kae Capital, Merck GHI Fund, Novo Holdings, TeamFund; Jan 2026 grant of $8M also reported. Compiled sources place total capital + grants at ~$129M (unaudited, varies by database). Private, venture/growth-backed; March 2022, HealthQuad and Novo Holdings led a $40M round for US/Europe expansion; stakes undisclosed.

DeepTek (India) — AI-enabled radiology platform: image viewing, workflow, reporting, analytics, chest X-ray/TB screening. Augmento (radiology-ops integration) and Genki (high-volume lung screening) product lines; integrates software with radiology workflow and, in some markets, teleradiology services. 2,000+ global sites, 3M+ lives touched, 7 patents filed (current figures); an earlier JICA profile cited 350+ paying hospitals/imaging centers, 55,000+ scans/month, ~5 Singapore public-sector hospital deployments — different dates, should not be combined as one current KPI set. March 2022: $10M Series A led by Tata Capital Healthcare Fund II with Pentathlon Ventures and GHV. A commercial database estimated FY2025 revenue at ₹35.1 crore — not a company disclosure, unverified. Private, PE/VC-backed; stakes undisclosed.

Gleamer / DeepHealth — cloud-based AI across X-ray, mammography, CT, MRI: fracture detection, lung-nodule analysis, MS lesion assessment. March 2026: Nasdaq-listed RadNet agreed to acquire Gleamer and integrate with its DeepHealth subsidiary. At announcement: 700+ customer contracts, 30M+ examinations analyzed across 2,250 locations in 44 countries, ~$30M expected 2026 ARR. RadNet agreed to pay €215M upfront plus up to €15M milestones (up to €230M total) for 100% acquisition — implies roughly 9x expected 2026 ARR before earn-out/growth/cash/debt adjustment (calculated from disclosed figures, not reported by the companies). Pending/completed strategic acquisition by RadNet, not a PE transaction.

Porter’s Five Forces

(Scale: 1–5, where 5 = strongest pressure on industry profitability)

PE Firms, Stakes, and Strategic Vision

Disclosed transactions are predominantly growth-equity and strategic-control deals, not traditional leveraged buyouts. Minority percentages are generally undisclosed. The sector’s negative-to-modest cash flow, high R&D expense, and continuing regulatory investment make it better suited to growth capital or a strategic add-on than a conventional high-leverage standalone buyout.

Regulatory Context

United States

• FDA regulates AI imaging products meeting the medical-device definition: 510(k) (substantial equivalence to predicate, currently the dominant route for radiology AI), De Novo (novel low-to-moderate-risk devices without a suitable predicate), PMA (higher-risk Class III devices). CDS software may fall inside or outside device regulation depending on functionality, intended use, and whether the clinician can independently review the recommendation basis.
• AI total-product-lifecycle draft guidance (Jan 6, 2025) proposed recommendations on design, development, documentation, transparency, bias mitigation, post-market performance monitoring — remains draft, representing FDA’s current thinking rather than a final binding rule.
• Final Predetermined Change Control Plan guidance (Aug 18, 2025): an authorized PCCP describing planned modifications, validation methodology, and impact may avoid a new marketing submission for every update; applies across 510(k), De Novo, PMA pathways. Investment implication: PCCPs reduce regulatory friction for planned updates but raise the quality threshold for validation, change management, bias analysis, cybersecurity, post-market surveillance — companies with mature regulatory/MLOps infrastructure should gain an advantage over single-model startups.

India

• CDSCO and Drugs Controller General of India regulate medical devices under the Drugs and Cosmetics Act, 1940 and Medical Devices Rules, 2017; state licensing authorities handle lower-risk categories, central licensing authority handles higher-risk/imported devices.
• ICMR’s ethical guidance for AI in biomedical research/healthcare addresses responsible development but is not a substitute for a CDSCO license.
• Most important recent change: CDSCO’s medical-device-software guidance (2025 draft, followed by 2026 guidance CDSCO/MD/GD/MDSW/01/2026) clarifies MDR 2017’s application to Software in/as a Medical Device — covers intended use, risk classification, applicable standards, QMS, licensing, clinical investigations, documentation, post-market requirements. Does not create a separate statutory framework; MDR 2017 remains controlling. For imaging vendors, standalone radiology AI intended to diagnose, screen, monitor, or support clinical decisions can qualify as SaMD and must be classified by intended use and clinical impact.
• Confidence flag: no reliable public total for CDSCO-authorized AI imaging products was found. Any target claiming “CDSCO approved” should provide the actual license, product classification, intended-use wording, manufacturing/import authorization, and evidence the marketed version matches the licensed version.

Risk Factors

1. Clearance without commercial adoption — FDA clearance establishes permission to market, not customer demand, reimbursement, clinical superiority, or workflow ROI. 183 of 221 identified manufacturers in the Innolitics analysis held only one clearance — a target can hold valuable regulatory assets but still fail on integration costs, sales cycles, or limited budgets. Diligence test: revenue and live volume by product, site activation rate, contract-to-production time, renewal rate, algorithm utilization, gross retention.
2. Point-solution commoditization and platform bundling — hospitals increasingly want enterprise platforms over separate per-finding contracts. Aidoc’s aiOS and RadNet’s DeepHealth portfolio (via Gleamer) both consolidate distribution; a single-indication company may lose pricing power or be relegated to a marketplace revenue share, since the platform controls distribution, workflow integration, and the customer relationship.
3. Model drift and poor external generalization — changes in patient demographics, disease prevalence, scanner vendors, imaging protocols, clinical practice can degrade post-deployment performance. Lower sensitivity creates missed findings and liability exposure; lower specificity creates alert fatigue and loss of clinician trust — either can cause a hospital to quietly disable the product without formally terminating the contract.
4. Weak unit economics despite high scan volume — India and public-health screening markets generate large clinical volumes but generally lower software revenue per scan than US health systems; cloud inference, local implementation, support, regulatory maintenance, tender discounts, and channel revenue sharing can consume gross margin. “Millions of scans processed” can coexist with low ACV and negative contribution margin — lives touched is not a revenue proxy.
5. Regulatory change and version-control burden — FDA’s PCCP can speed planned updates, but changes outside an authorized PCCP may still require a new submission; India’s software guidance similarly raises QMS, evidence, licensing, post-market-control expectations. A company unable to prove which model version ran at each customer, how it was validated, and whether use stayed within the authorized intended use may face deployment freezes or remediation costs.

PE Attractiveness Verdict

Selective. The most commercially validated clinical-AI sub-sector — hundreds of annual FDA clearances, enterprise deployments across major health systems, several vendors at tens of millions of patients/examinations per year. However, regulatory clearance alone provides little moat, point-solution rivalry is extreme, buyers hold substantial power, India market-size estimates are unreliable, and workflow platforms are consolidating the value chain. The most attractive targets are not isolated algorithms but businesses with multi-product regulatory portfolios, embedded PACS/RIS integrations, demonstrable recurring revenue, high live-site utilization, external clinical validation, robust post-market monitoring, and a credible path to owning the reporting or care-coordination workflow. For conventional PE, the preferred entry is a scaled platform or a profitable radiology-services/imaging asset to which AI can be added as a productivity/differentiation layer — earlier point-algorithm companies are better treated as selective growth investments or strategic add-ons than standalone leveraged-buyout platforms.