How Indian Hospitals Are Using AI in Diagnostics: What Doctors Need to Know

Three Indian AI diagnostics companies have moved from startup to scaled clinical deployment in the past three years: Niramai (breast cancer screening via thermography), Qure.ai (chest X-ray analysis for tuberculosis and pneumonia), and Tricog (real-time ECG analysis for cardiac events). Each has published validation data. Each operates within India's evolving regulatory framework. Each is deployed at meaningful scale - not in pilot phase.

What most physicians and hospital administrators do not know is what these tools actually do, what they do not do, and what clinical, regulatory, and integration obligations come with adopting them. The gap between vendor marketing and clinical reality is wide enough to cause genuine harm if it is not understood before adoption.

This article covers the verified record on each platform, India's regulatory framework for AI as a medical device (SaMD) under CDSCO and ABDM, and a practical framework - the VERIFY method - for evaluating any AI diagnostic tool before your hospital adopts it. After reading this, you will know the right questions to ask, the answers that should concern you, and the red flags that mean slow down.

---

At the cardiology department of a mid-sized private hospital in Nashik, a 54-year-old patient presents at 11 PM with atypical chest pain. The on-call resident orders an ECG. The duty cardiologist is 40 minutes away. Three years ago, that ECG would sit on a machine until someone read it. Tonight, the hospital uses Tricog - the ECG image is transmitted, analyzed by AI within 60 seconds, and a trained cardiologist reviews the interpretation remotely. The STEMI alert changes the care pathway before the on-site cardiologist even arrives.

This scenario is technically accurate in its mechanics. Across India, similar situations are playing out in radiology departments where Qure.ai's qXR is flagging chest X-rays for TB-related findings before the radiologist opens the morning queue, and in mobile screening camps where Niramai's thermal imaging system is running breast cancer risk assessments in communities that have no mammography machine within 200 kilometres.

The technology works - with specific caveats. And those caveats matter enormously for any physician or administrator considering adoption. The three most common mistakes hospitals make when evaluating AI diagnostics tools are: (a) accepting vendor-provided accuracy figures without examining the validation study population; (b) assuming CE marking or US FDA clearance transfers directly to Indian clinical and regulatory compliance; and (c) underestimating the clinical governance requirements that remain with the institution, not the vendor, after deployment.

A fourth mistake is more subtle: treating AI diagnostic tools as a cost-reduction mechanism without building the clinical oversight infrastructure that the technology actually requires. When an AI tool flags a chest X-ray as abnormal, a qualified human must still act on that flag. The workflow does not disappear. It changes shape - sometimes for the better, sometimes in ways that introduce new error modes that did not exist in the prior process.

The 2025 FICCI-EY Parthenon survey of over 1,000 patients and 100 clinicians noted that patients increasingly rely on informal proxies like brand reputation and word-of-mouth to assess care quality - not clinical outcomes. That dynamic puts pressure on hospitals to deploy visible, legible technology. But patient-facing visibility and clinical validity are not the same standard. Conflating them is how AI diagnostics tools get adopted for the wrong reasons, with insufficient governance in place when something goes wrong.

---

India's healthcare infrastructure creates both the largest opportunity and the most specific constraints for AI diagnostics. The country has approximately one radiologist for every 100,000 population - compared to ten per 100,000 in the United Kingdom. Tuberculosis remains the leading infectious disease killer. Cardiovascular disease causes an estimated 28% of all deaths. Breast cancer is the most common cancer among Indian women, and it is diagnosed at a median stage later than in high-income countries, partly because of access gaps in screening infrastructure. These are the specific conditions AI diagnostics tools have been built to address in the Indian context - and the fit between product design and India's disease burden is not incidental.

India crossed 800 million smartphone users in 2024, and the median age is 29. The infrastructure base for cloud-connected diagnostic devices - tablets for thermography image capture, smartphones for ECG transmission, digital X-ray systems for AI-assisted reads - is broadly present even in secondary cities and some rural settings. This is why deployment at scale is possible in ways that would have been technically implausible five years ago. The connectivity and device penetration that makes digital patient acquisition viable for hospitals also makes AI-assisted diagnostics operationally feasible beyond tier-1 cities.

The regulatory landscape, however, is in active transition. CDSCO (Central Drugs Standard Control Organisation) brought AI-based software as medical devices (SaMD) under the Medical Devices Rules framework, starting with the 2017 rules and subsequent amendments. In practice, this means AI diagnostic tools are classified as Class B or Class C medical devices depending on risk level, require CDSCO registration, and are subject to post-market surveillance requirements. A meaningful number of tools currently marketed to Indian hospitals had not completed CDSCO registration as of 2024 - some because they qualify for registration but the process was still underway, some because the regulatory interpretation of their specific category was actively contested.

ABDM (Ayushman Bharat Digital Mission) adds a parallel but distinct framework. ABDM defines standards for health data exchange, establishes ABHA (Ayushman Bharat Health Account) as the patient identity layer, and creates data localization requirements relevant to AI tools that transmit patient data to cloud infrastructure. Any AI diagnostic tool that stores, processes, or transmits patient health data is also subject to the Digital Personal Data Protection Act 2023 (DPDPA), which came into full operational effect in 2025. The DPDPA introduces consent obligations and data principal rights that have direct operational implications for how hospitals collect, store, and share patient imaging and diagnostic data. Hospitals that have not updated patient consent frameworks to cover AI-assisted analysis are carrying DPDPA exposure they may not have assessed.

---

Qure.ai's qXR platform has been the most extensively published of the three Indian AI diagnostics companies in this category. A 2021 prospective validation study published in PLOS Medicine, involving external clinical evaluation, assessed qXR against radiologist reads for TB-related findings across a large multi-site dataset drawn from India. The study reported area under the curve (AUC) figures above 0.90 for multiple TB-relevant finding types. Methodologically, the study used Indian X-ray data - which partially addresses the critical validation population question - and was conducted at multiple sites rather than a single referral centre. Limitations include the retrospective elements in dataset curation and the concentration of evaluation sites in larger urban centres. A subsequent operational deployment study in partnership with the Stop TB Partnership and government TB programs reported real-world performance data from lower-resource settings across multiple countries, including India. Label: Indian data, peer-reviewed; meaningful validation, but generalizability to rural district-level equipment requires further evidence.

Tricog's InstaECG platform has published outcomes data from real-world deployment across Indian hospitals. A 2020 observational study published in Indian Heart Journal reported door-to-balloon times in Tricog-integrated hospitals compared to non-integrated comparators, demonstrating clinically significant reductions for STEMI patients. The study design was observational and non-randomised - standard limitations regarding confounding and selection bias apply. Critically, Tricog's platform combines AI ECG analysis with human cardiologist review in a hybrid model. This is a materially different risk profile from a fully autonomous AI diagnostic system: the human cardiologist review is not optional under Tricog's standard deployment model, and this architectural decision substantially changes the clinical governance calculus for hospitals that adopt it. Label: Indian real-world data, peer-reviewed, observational design; hybrid AI-plus-human model is a meaningful risk distinction.

Niramai's SMILE (Smart Medical Imaging and Learning Engine) platform has published feasibility and accuracy data in clinical and technical publications. A validation study in Breast Cancer Research and Treatment (Rangarajan et al.) reported sensitivity and specificity figures for the thermography-plus-AI approach in an Indian patient cohort. The population was drawn from a single urban tertiary centre, which limits generalizability to the rural outreach settings where Niramai's deployment model is often positioned. It is essential to note that thermography-based breast cancer screening is not equivalent to mammography. The evidence base for thermography as a standalone screening modality is substantially weaker than for mammography, and no major clinical guideline recommends thermography as a first-line screening tool where mammography is available. Niramai's clinical value proposition is access in mammography-unavailable settings, not superiority over mammography where it exists. Label: Indian data, peer-reviewed; single-site validation cohort; thermography is not a mammography substitute - the use case is access-gap screening.

US and global evidence on radiology AI is extensive and directionally applicable to India with important caveats. A 2023 systematic review in The Lancet Digital Health covering 69 AI diagnostic studies found consistent accuracy in controlled research settings but mixed real-world deployment performance. Several studies documented AI performance degradation when tools were deployed on equipment or patient populations different from training data. This finding is directly and specifically relevant to Indian deployment contexts: AI tools trained primarily on Western patient data - different X-ray equipment manufacturers, different TB prevalence patterns, different population body habitus distributions, different image acquisition protocols - may show reduced accuracy in Indian clinical environments. This is not a hypothetical risk. It is a documented mechanism. Label: Global evidence, directionally applicable to India; equipment and population heterogeneity is the primary caution for Indian adoption.

On regulatory compliance and clinical outcomes, there is a documented gap in the published literature. No peer-reviewed study has examined clinical outcomes specifically attributable to CDSCO-compliant versus non-compliant AI diagnostic tool deployment in India. The regulatory framework exists and carries legal force. Outcomes data stratified by compliance status does not exist in the published record. Hospitals should not interpret the absence of documented regulatory harm as evidence that compliance is optional. Label: observed evidentiary gap; regulatory obligation exists independent of outcomes data.

---

The following section is Influx Health's interpretation and opinion, not research.

Working with over 60 Indian healthcare organizations, we have observed a consistent pattern in AI diagnostics adoption: hospitals that deploy these tools for genuine operational reasons - covering radiologist shortfalls, improving ECG triage speed, extending screening reach - achieve more durable clinical integration than hospitals that adopt primarily for marketing differentiation. When the use case is operationally real, workflow design tends to be more careful. When the use case is primarily brand-facing - "we have AI" - clinical governance is treated as an afterthought and often never gets built properly.

A specific observation worth flagging: the CDSCO registration question is routinely not asked during vendor evaluation. We have seen hospitals complete full procurement processes for AI diagnostic tools without requesting the CDSCO registration certificate. When we raise the question, the most common response from hospital administrators is genuine surprise - they did not know the requirement existed or assumed a government program endorsement covered it. This is not a regulatory technicality. A CDSCO-unregistered AI tool used in clinical decision-making creates liability exposure for the hospital and potentially the operating physician, not the vendor. The standard vendor contract language almost never assigns this liability clearly. When a clinical adverse event occurs, the contract language becomes very important very quickly.

The second pattern we observe concerns how validation population claims are presented. Vendors consistently lead with aggregate accuracy figures - "95% sensitivity for TB findings" or "AUC of 0.93" - without volunteering the detail of which dataset those figures were derived from. When pressed, a meaningful proportion of published validation studies turn out to be primarily on non-Indian datasets, or on Indian datasets from a narrow urban demographic or specific high-end equipment profile. An AI tool validated on digital X-rays from a modern imaging system in a Mumbai tertiary referral centre may perform differently on X-rays from a lower-resolution machine in a district hospital in Jharkhand or Chhattisgarh. This is a known risk in AI medical device deployment globally. Indian hospitals should ask about it every time, without exception.

Third, and perhaps most practically important: the clinical oversight requirement does not disappear when AI is deployed. Tricog's platform explicitly builds human review into the service architecture. But for AI tools that produce a risk score or flag without a mandatory human review layer embedded in the product, the hospital must build that oversight into its own workflow. We have observed situations where AI-generated flags were operationally deprioritized because the volume of flags exceeded the available physician review capacity. That is a workflow design failure, not an AI failure - but the clinical consequence to the patient is identical. The hospital designed a system it could not operate at the required quality level, and it is the hospital's responsibility, not the vendor's.

---

Before adopting any AI diagnostic tool, hospital administrators and clinical leads should work through each element of the VERIFY framework. This is not a vendor scorecard. It is a clinical governance and institutional risk assessment tool.

V - Validation population: Was the tool validated on Indian patient data? Demand the primary validation study, not the product sheet. Check whether the study population was drawn from India, from which regions, from which equipment manufacturers, and from which demographic profile. A tool validated on US or European hospital data carries unknown accuracy when deployed on Indian patients. Ask specifically: "What percentage of your validation data was from Indian patients, on Indian equipment, and across both urban and rural settings?" If the vendor cannot answer this question in specific terms, the tool has not been adequately validated for Indian clinical deployment.

E - Evidence quality: What is the study design behind the accuracy claim? Retrospective studies on curated datasets consistently outperform prospective real-world deployment in published AI diagnostics literature. A sensitivity figure from a controlled retrospective study is not the same as real-world clinical performance at your hospital. Ask for prospective validation data or real-world deployment outcomes data from settings comparable to yours. If neither exists, the tool is in de facto pilot phase regardless of how it is marketed. You are the pilot.

R - Regulatory status: Is the tool registered with CDSCO as a medical device? Request the CDSCO registration certificate and the device classification. Confirm whether the specific clinical use case described in the vendor's marketing materials matches the approved indication exactly. Some tools have CDSCO registration for a narrower indication than they are marketed for. Also note whether the tool carries CE marking or FDA clearance - these are not substitutes for CDSCO registration in India, but their presence indicates the vendor has undergone rigorous third-party regulatory evaluation in other jurisdictions and has the infrastructure to manage regulatory processes.

I - Integration requirements: What does clinical implementation actually require? AI diagnostic tools do not deploy into a vacuum. Map the full integration requirement: connectivity to your Hospital Information System or EMR, DICOM compatibility for imaging tools, minimum network bandwidth and reliability requirements, data storage location (cloud versus on-premise, and in which jurisdiction), and DPDPA-compliant patient consent flow integration. Any tool that stores or transmits patient health data must be covered in your DPDPA-compliant consent architecture. From 2025 onward, this is a legal obligation, not a best practice.

F - Frontline oversight: Who is clinically responsible for acting on AI outputs? Define in writing, before deployment, who is responsible for reviewing AI-generated flags or scores, within what timeframe, with what documentation requirement, and what happens when the AI output is not acted on within that timeframe. The AI tool itself carries no clinical liability. The hospital and the operating physician do. Tricog's hybrid model embeds human cardiologist review into the service; for tools that do not include a mandatory human review layer, the hospital must build equivalent oversight into its own workflow. Failure to do so is not only a regulatory risk - it is a patient safety risk that will eventually materialize.

Y - Yield calculation: What is the actual cost-benefit over the full deployment lifecycle? Model the total cost of ownership over 36 months: licensing or per-study fees multiplied by realistic monthly volume, hardware requirements, staff training and re-training, integration development costs, and ongoing clinical governance overhead. Then model the realistic benefit using conservative assumptions: radiologist time savings (quantified against your actual radiologist cost structure), downstream clinical outcome improvements (use published data, not vendor projections), and any quality indicator improvements reportable to NABH or other accreditation bodies. If the yield calculation does not hold at 36 months under conservative assumptions, the deployment economics are not sound regardless of the technology's clinical merit.

---

• Request the CDSCO registration certificate from every AI diagnostics vendor currently under evaluation or in active use. If a vendor cannot produce current registration documentation within 48 hours, pause procurement or deployment until regulatory status is fully clarified in writing and confirmed by your legal counsel.
• Pull the primary peer-reviewed validation study - not the product brochure - for any AI tool currently deployed or under evaluation. Check the study population size, demographic profile, equipment manufacturer, and geographic origin. Record this in your procurement file and have a clinical lead review it before sign-off.
• Audit your patient consent forms for DPDPA coverage. If current consent forms do not explicitly cover AI-assisted analysis and cloud transmission of imaging data, update them before the next contract renewal or new deployment. This is a 2025 legal obligation.
• Designate a named physician as clinical lead for each AI diagnostic tool in use. This person is responsible for reviewing flagged cases within a defined SLA, logging adverse events or near-misses, and escalating governance concerns. This designation should be in writing and reflected in your clinical governance documentation.
• Run a 36-month cost of ownership calculation for any AI tool in active procurement, using per-study fee estimates multiplied by realistic monthly volume plus integration, training, and governance costs. Compare the result against the specific operational bottleneck the tool is meant to address.
• If you are considering Niramai for rural outreach or screening programs, confirm the full operational pathway for elevated-risk findings before deployment - specifically, the referral pathway to confirmatory mammography or biopsy, transport support if required, and the follow-up tracking mechanism. An AI screening tool without a confirmed onward-referral pathway generates patient anxiety without clinical value.

---

Q: Qure.ai and Tricog are mentioned in government health programs. Does government endorsement mean CDSCO approval?

Not automatically. Government health programs - including those under the National Health Mission and the Stop TB Partnership - have partnered with AI diagnostics vendors for specific operational use cases. These partnerships reflect a procurement and public health operational decision, not a regulatory certification or endorsement of CDSCO compliance. CDSCO registration is a separate process administered by the Central Drugs Standard Control Organisation under the Medical Devices Rules. It is entirely possible for a tool to be used in a government program while its SaMD registration is pending, operating under a specific exemption, or contested. Always verify CDSCO registration status directly and independently. Request the certificate, not a verbal assurance.

Q: Our hospital is NABH accredited. Does NABH have specific requirements for AI diagnostic tools?

As of mid-2026, NABH does not yet have a dedicated chapter specifically governing AI diagnostic tool adoption and oversight, though this is expected to change in the forthcoming NABH 6th Edition revision cycle. However, existing NABH standards on clinical decision support systems, diagnostic quality assurance, and medical device management are applicable to AI diagnostic tools deployed in a NABH-accredited setting. Specifically: device calibration and maintenance documentation, incident reporting processes for diagnostic errors (including AI-influenced ones), and clinical governance documentation requirements all apply. Document your AI oversight protocols to these existing standards now, and maintain that documentation in a form auditors can review.

Q: A vendor says their tool has FDA 510(k) clearance. Is that sufficient for clinical use in India?

No. FDA 510(k) clearance indicates the tool has undergone rigorous evaluation by the United States Food and Drug Administration - this is relevant context and a positive indicator of regulatory maturity. But it is not a substitute for CDSCO registration in India. Indian medical device regulatory requirements are separate and must be met independently. Additionally, FDA clearance is specific to an indication: confirm that the Indian clinical use case matches the cleared indication exactly. A tool cleared as a clinical decision support aid is not automatically cleared as a standalone diagnostic device. The same principle applies to CE marking under EU MDR - valuable evidence of regulatory process maturity, not an Indian regulatory substitute.

Q: We currently use Qure.ai for chest X-ray triage. What ongoing governance obligations do we have?

At minimum: maintain documentation confirming that all AI-flagged studies are reviewed by a qualified radiologist or physician before clinical action is taken; log and investigate any case in which an AI output was relied upon without adequate human review and a clinical adverse event or near-miss occurred; maintain records related to any CDSCO-registered medical device in active use as required under the Medical Devices Rules; and confirm that patient consent covers AI-assisted analysis under your DPDPA framework. Review Qure.ai's data processing agreement - specifically, confirm where patient imaging data is stored and processed, the jurisdiction of that storage, and whether this is consistent with your DPDPA and ABDM obligations.

Q: Niramai uses thermography rather than mammography. Should we be concerned about efficacy for clinical decisions?

Yes, with important qualification. Thermography-based AI is not validated as a standalone breast cancer diagnostic tool, and no major clinical guideline recommends it as a first-line screening modality where mammography is available. Niramai's clinical value proposition is access in mammography-unavailable settings - communities where the alternative to thermography screening is no screening, not mammography. If your facility has operational mammography capacity, Niramai is not a replacement for it. If you are deploying for outreach to populations that genuinely lack mammography access, the critical governance question before deployment is: what is the confirmed clinical pathway for a woman who receives an elevated risk score? Without a mapped referral pathway to confirmatory imaging and biopsy - including transport support if needed - AI screening in resource-limited settings generates patient anxiety without delivering the clinical benefit that is the entire justification for the program.

---

Internal - Influx Health Institute - What Is ABDM and Why Every Indian Hospital Needs to Understand It - Telemedicine Regulations in India: A Practical Guide for Private Hospitals - Building Patient Trust in the Digital-First Healthcare Era

External - Authoritative Sources - CDSCO Medical Devices and Diagnostics Division - regulatory framework, SaMD classification, and registration requirements: https://cdsco.gov.in/opencms/opencms/en/Medical-Device-Diagnostics/Medical-Device-Diagnostics/ - Ayushman Bharat Digital Mission (ABDM) - ABHA framework, health data exchange standards, and SaMD governance: https://abdm.gov.in - The Lancet Digital Health - peer-reviewed AI diagnostics research including systematic reviews on real-world AI performance: https://www.thelancet.com/journals/landig/home

---

Read Next: What Is ABDM and Why Every Indian Hospital Needs to Understand It - the regulatory and data infrastructure context behind AI diagnostics in India, including SaMD governance, ABHA integration requirements, and what DPDPA means for patient data in hospital AI systems.

Assess Your Practice: Run the Digital Presence Meter at influx-health.com/dpm - the same rigorous, evidence-first discipline that governs AI tool evaluation applies to your hospital's patient-facing digital presence. See where you stand.

Chat with Influx Health: If you are evaluating AI diagnostics adoption and want a direct conversation about vendor evaluation, governance frameworks, or what questions to ask before you sign, speak with the Influx Health team at influx-health.com/contact.

--- ---

# Content Derivatives: Center 7, Article 3

---

Subject line: The AI diagnostics question your vendor evaluation process hasn't asked yet

---

Dear Dr. [Name],

Three Indian AI diagnostics companies have moved from startup to meaningful clinical deployment in the past three years. You have likely heard of Qure.ai, Tricog, and Niramai. What is less commonly discussed is what your hospital's actual obligations are when you adopt one of these tools - and what questions most procurement processes never ask.

The most important question is also the one most frequently skipped: was this tool validated on Indian patient data? Accuracy figures cited in vendor materials are often derived from curated retrospective datasets that may not reflect your patient population, your imaging equipment, or your clinical setting. A tool that performs at 95% sensitivity in a controlled validation study on a Mumbai tertiary centre dataset may perform differently on district hospital equipment with a different patient demographic. This is a documented risk in AI diagnostics globally, and it is specifically relevant in India given the heterogeneity of clinical settings.

The second question that should precede any adoption: is this tool registered with CDSCO as Software as a Medical Device (SaMD)? Many tools currently marketed to Indian hospitals are not fully CDSCO-registered. Government program deployment is not a substitute for CDSCO certification. When a clinical adverse event involves an unregistered medical device, the liability sits with the hospital and the operating physician - not the vendor. The contract language almost never makes this clear upfront.

The third question is about clinical oversight. AI tools produce outputs. Qualified clinicians must act on those outputs within defined timeframes. That clinical workflow does not disappear when you deploy AI - it changes shape. If you have not designed that workflow before deployment, you have deployed without a complete implementation.

The new article on the Influx Health Institute covers all three companies in clinical detail, India's regulatory framework under CDSCO and ABDM, and the VERIFY framework - a six-step evaluation checklist for any AI diagnostic tool your hospital is considering or currently using.

[Read the full article at influx-health.com/institute/center-7-ai-healthcare/ai-diagnostics-india]

[Run the Digital Presence Meter at influx-health.com/dpm]

---

AI in Indian diagnostics: what to know before your hospital adopts (12-minute read)

Three Indian AI diagnostic tools are now deployed at scale - not pilot phase. What each actually does:

1. Qure.ai (qXR) - chest X-ray AI flagging TB, pneumonia, lung nodules and 15+ findings. Validated on Indian datasets, published in peer-reviewed journals. A triage and prioritization tool, not a radiologist replacement.
2. Tricog - real-time ECG transmission + AI + cardiologist review for STEMI detection. Hybrid model: human review is built into the service, not optional.
3. Niramai - thermography AI for breast cancer screening in settings without mammography access. Not equivalent to mammography. Value is access-in-the-gap.

Before adopting any AI diagnostic tool, get answers to six questions:

1. Was it validated on Indian patient data - which region, which equipment?
2. What is the study design - retrospective curated data or real-world outcomes?
3. Is it CDSCO-registered as a medical device (SaMD)?
4. What are the integration requirements - HIS, DPDPA consent flow, data storage location?
5. Who is clinically responsible for acting on AI outputs, within what timeframe?
6. What is the 36-month total cost of ownership against the operational problem you're solving?

Government program use does not equal CDSCO approval. CE/FDA clearance does not substitute for Indian registration. DPDPA consent obligations apply to all patient data transmitted to cloud AI systems.

Full article: influx-health.com/institute/center-7-ai-healthcare/ai-diagnostics-india

---

Indian hospitals are adopting AI diagnostic tools faster than their clinical governance frameworks are being updated to manage them.

Three companies - Qure.ai, Tricog, and Niramai - have moved from startup to meaningful clinical deployment in the past three years. Each has published validation data. Each is solving a real problem: radiologist shortfalls for chest X-ray triage, ECG interpretation delays in hospitals without on-site cardiologists around the clock, breast cancer screening access in communities without mammography infrastructure. The underlying clinical need is real. The technology is real.

What the vendor conversations rarely cover: the regulatory and clinical governance obligations that come with adoption.

Most hospital procurement processes for AI diagnostic tools do not request the CDSCO registration certificate. Government program deployment is not a substitute for CDSCO registration. When a clinical adverse event involves an unregistered AI-based medical device, the liability sits with the hospital and the operating physician.

Most procurement processes do not verify the validation study population. A tool validated on curated datasets from urban tertiary centres may perform differently on the equipment and patient demographics of a secondary city hospital or district facility. This is a documented mechanism in AI diagnostics globally - not a theoretical concern.

Most procurement processes do not build clinical oversight workflows before deployment. AI tools produce outputs that qualified clinicians must act on within defined timeframes. That responsibility does not transfer to the vendor.

We have published a six-step evaluation framework - the VERIFY method - for assessing any AI diagnostic tool before adoption: Validation population, Evidence quality, Regulatory status, Integration requirements, Frontline oversight design, and Yield calculation.

Practical, specific, and not a vendor endorsement of any of the three companies discussed.

Read it at the Influx Health Institute. Link in comments.

---

1/ Indian hospitals are deploying AI diagnostics at scale. Three companies you should understand precisely before adoption: Qure.ai, Tricog, Niramai. What they actually do, what they don't do, and what your governance obligations are. A thread.

2/ Qure.ai (qXR): AI analysis of chest X-rays for TB, pneumonia, lung nodules, and 15+ other findings. Validation studies on Indian datasets published in peer-reviewed journals. Deployed in government TB programs. A triage and prioritization tool - not a radiologist replacement.

3/ Tricog: real-time ECG image transmission + AI analysis + human cardiologist review for cardiac events including STEMI. The cardiologist review is built into the service - it's not optional. Published outcomes data showing reduced door-to-balloon times in Indian hospitals.

4/ Niramai: thermography-based breast cancer screening using AI risk scoring. Designed for settings without mammography access. Not equivalent to mammography. The value is access-in-the-gap - but only if there is a confirmed referral pathway for elevated-risk findings.

5/ The question most procurement teams skip: was this tool validated on Indian patient data, from Indian equipment, across the demographic profile of your actual patient population? AI performance degrades when deployed on populations different from training data. This is documented.

6/ CDSCO: AI-based diagnostic tools are classified as Software as a Medical Device under Indian Medical Devices Rules. CDSCO registration is required. Government program deployment is not a substitute. Clinical adverse events involving unregistered devices = liability on the hospital.

7/ DPDPA 2023: any AI tool that transmits patient imaging or diagnostic data to cloud infrastructure must be covered in your DPDPA-compliant patient consent framework. In effect since 2025. Check your current consent forms - most haven't been updated.

8/ Frontline oversight: AI tools produce outputs. Qualified clinicians must act on those outputs within defined timeframes. That workflow does not disappear when you deploy AI. It changes shape. Design the oversight workflow before deployment, not after the first incident.

9/ The VERIFY framework for evaluating any AI diagnostic tool before adoption: Validation population → Evidence quality → Regulatory status → Integration requirements → Frontline oversight → Yield calculation. Each step is a concrete question that demands a concrete answer from the vendor.

10/ Full article with clinical detail on all three platforms, India's CDSCO and ABDM regulatory framework, and the complete VERIFY framework: influx-health.com/institute/center-7-ai-healthcare/ai-diagnostics-india

Assess your hospital's digital presence: influx-health.com/dpm

--- Article published by the Influx Health Institute. Influx Health is a patient acquisition agency for healthcare organizations in India.