India has roughly 22 crore women at risk of breast cancer. It screens fewer than 2% of them in a year — about 60 lakh. The gap is not only one of reach and cost; it is also one of measurement. The artificial intelligence that powers most pathology screening was trained on Western tissue, and on India's 9.2 crore dense-breast women it reads at 60–70% accuracy. DigiClinics is built on the argument that closing that gap needs AI trained on Indian morphology in the first place, paired with a device cheap enough to sit in a primary health centre.
The company is based in Hyderabad. The two co-founders, Dr. Raj Subramanian and Dr. Devika Rubi Rajasekaran, met as post-doctoral fellows in computational pathology at Emory University in Atlanta between 2014 and 2017, where they co-authored the pathomics framework that now underlies the company's lead product. Since 2017 they have built what the deck describes as a sovereign AI diagnostics platform, assembling a 154 terabyte Indian whole-slide-image dataset and over 1.1 million pathologist-annotated tiles across eight years of continuous research.
What follows is our read on the proposition — the problem, the two-part product, the founders, the market, the competitive position and the watchpoints.
A screening system that does not reach, or read
The deck frames the failure in two halves: the old format and the old technology. On format, camp-based screening fails on last-mile economics — stigma, travel and privacy keep women away — while mammography sensitivity falls to 30–50% on the 9.2 crore women with dense breast tissue. Late-stage detection costs around 45% more than early-stage, and survival at Stage III or IV sits below 30%.
On technology, the constraints compound. Imported portable ultrasound runs ₹15–30 lakh, out of reach for the 40,000-plus primary and community health centres. Pathology backlogs run six to eight weeks against a ratio of one pathologist per 1.5 lakh population. And Western-trained AI clocks 60–70% on Indian morphology, with no PNDT hardening built in anywhere — a serious gap in a country where prenatal sex determination is illegal and ultrasound devices are tightly regulated.
The cost of all this, per the deck, is around USD 1.5 billion a year in avoidable late-treatment spend, plus roughly ₹50 lakh per family in productivity loss for each preventable death. The deeper structural diagnosis is fragmentation: single-cancer point tools that rebuild the full stack for every disease site, Western-trained models whose Indian errors hide in the tail of the distribution, and hardware-software siloes where imaging AI, PACS, EHR and LIS are sold separately with no integrated workflow.
Two halves: a co-pilot and a device
DigiClinics is building two things designed to work as a tandem. OncoSynopticAI is the pathology and radiology co-pilot. BharatScan iUSG is the indigenous ultrasound. The pitch is a single screen-read-decide workflow at the health centre: BharatScan captures the image and runs AI BI-RADS scoring at the PHC or CHC; OncoSynopticAI pre-fills CAP and AJCC-8 synoptic reporting fields from the whole-slide image; and decision-grade biomarkers feed therapy choice and a tele-oncology workflow.
The numbers the company attaches to that workflow are specific. AI accuracy of 94–97% on Asian-Indian cohorts against 60–70% on Western-trained models. A 70–85% concordance improvement in Nottingham grading versus an unaided pathologist read. An 18–24% sensitivity uplift in dense-breast ultrasound BI-RADS scoring. And a pathology turnaround that drops from 14 days to 48 hours, with pathologist review time per case compressed from 45–60 minutes to under 15.
The model keeps a human in the loop. The pathologist still signs out — the AI pre-fills the synoptic fields and enforces completeness before the report can close.
— On the co-pilot designThe biomarker accuracies are the heart of the clinical claim: mitosis at 94%, ER/PR at 97%, HER2 at 94% and sentinel lymph node at 94%, with two of the pipelines (ER/PR and HER2) described as patented. The device half is positioned as a world-first on two fronts: PNDT hardening built into hardware and firmware — region-locked presets, fetal-detection auto-disable, geofencing and an audit log — and AI-guided self-screening, where a woman can take a private scan herself, which the company argues reduces stigma and lifts participation.
Clinical-AI researchers, with a lab network behind them
Dr. Raj Subramanian
Originated the whole-slide-image analysis architecture for breast cancer during a post-doctoral fellowship in computational pathology at Emory University. Reports 26-plus years across MedTech, healthcare imaging and clinical AI delivery, and is PMP certified. Has led the build of the 154 TB Indian dataset and 1.1M+ annotated tiles since 2017.
His co-founder, Dr. Devika Rubi Rajasekaran, leads AI imaging and genomics as CSO and is the CNN-plus-ViT architecture lead. The deck credits her with owning the Asian-Indian morphology training that produces the 94–97% accuracy figure, and she co-authored the underlying pathomics framework alongside Dr. Raj at Emory. The clinical side is anchored by Dr. Rohit Tapadia, Chief Clinical Officer and founder of a respected Hyderabad pathology lab, who serves as onco-pathology lead and clinical validation principal investigator. The advisory bench includes clinicians from Basavatarakam IACH, Osmania, AIIMS Bibinagar and Tapadia.
India is the validation rail; the US and EU are the value rail
The global AI-in-oncology market is sized at USD 6.0 billion today, growing to USD 38 billion by 2033 — a 24.8% compound rate — with diagnostics the largest application segment and breast the largest cancer-type segment. The deck breaks the serviceable market down by cancer site and geography, putting the cross-site addressable market in India at roughly ₹9,700 crore, with another USD 2,820 million across the Middle East and Southeast Asia and USD 15,950 million across the US and EU.
Breast is the only live segment today, with an India SAM of ₹3,500 crore. Lung (₹2,800 crore, trial-stage) and prostate (₹1,400 crore, validation) follow, with cervix-plus-HPV and oral head-and-neck on the roadmap. The strategic framing is that India-trained AI plus US or EU clearance is what global pharma and CROs will pay USD 1–3 million per companion-diagnostics engagement for — India supplies the validation cohorts, the developed markets supply the value.
The case rests heavily on policy timing. The deck lists six independent regulatory and programmatic moves converging on the screening problem: the National Health Mission's cancer-screening expansion mandating breast, cervical and oral screening through the PHC/CHC network; Ayushman Bharat's 1.5 lakh health and wellness centres; a formalised CDSCO software-as-a-medical-device pathway that parallels the FDA's 510(k); the ABDM data-exchange rail for federated records; the ICMR-DHR national AI portal backing population-calibrated datasets; and a MeitY indigenous-MedTech push, including a C-MET Thrissur transducer aimed at substituting around 90% of import dependency at a sub-₹5 lakh price.
The only player combining all three pieces
The deck divides rivals into three camps. National peers — SigTuple, Qritive, Morphle, Aindra and iOncology.ai — are described as hematology, cervical or workflow-only, without breast pathology grading, CAP/AJCC-8 synoptic automation, a diagnostics-platform-as-a-service model, or a full IHC and sentinel-lymph-node pipeline. International pathology AI — Paige.AI, Indica HALO AP and PathAI — carries Western-centric training sets, the 60–70% accuracy problem on Indian cohorts, and no PHC/CHC deployment model. Global portable ultrasound — GE Vscan Air, Philips Lumify, Butterfly iQ+ — sits at a ₹15–25 lakh price point or depends on the cloud, runs Western AI, and offers no PNDT compliance or self-screening.
DigiClinics positions itself as the only integrated player combining India-trained pathology AI (1.1M+ tiles, kappa above 0.78), CAP/AJCC-8 synoptic automation delivered as a service without scanner capex, and a PNDT-hardened indigenous ultrasound at ₹3–5 lakh. The claim is that no comparable global product combines all three — which, if it holds, is the company's core defensibility argument.
The dataset
154 TB India-trained whole-slide images and 1.1M+ annotated tiles. The deck argues this is not replicable in 12–18 months — the data, not the model, is the barrier.
The IP
Two patented biomarker pipelines (ER/PR, HER2). CDSCO SaMD plus India-residency posture is positioned to close the regulatory arbitrage against Western incumbents.
Distribution
The NHM and Ayushman Bharat channel — 40,000+ PHCs/CHCs and 1.5 L HWCs — described as the biggest moat, reaching population scale at zero hospital capex.
What is already on the board
The pre-Series A achievements the deck reports include a BIRAC BIG-16 government deep-tech grant, with a four-GPU NVIDIA A100 cluster commissioned via KMIT and ₹3.5 crore of infrastructure; a TIH-IIT Patna equity investment at 2.78% alongside ₹50 lakh of founder self-funding, for ₹5.5 crore total deployed; the Nottingham trial results (mitosis 94%, ER/PR 97%, HER2 94%, SLN 94%) confirming TRL 6–7; clinical anchors including NCG-India, Tata Memorial, AIIMS, Rajiv Gandhi CI, Fortis, Mayo Clinic and Harvard BWH; and a BharatScan iUSG proposal shortlisted at ANRF-MahaMedTech (Gates Foundation), the Technology Development Board and DoP-PRIP.
On data protection, the company describes a sovereign-data architecture spanning three regimes: India's DPDP Act 2023 (with all Indian patient data on India-pinned servers in the Hyderabad and Chennai region, and a penalty cap of up to ₹250 crore per breach noted as a board-level risk); HIPAA plus FDA SaMD rules in the US; and GDPR, the EU AI Act and MDR/IVDR in Europe. The stated operating posture is that India data stays in India, US data in the US and EU data in the EU, with only de-identified model weights — never patient images — crossing borders.
What the next twelve months will decide
The plan is sequenced around validation first, then commercial rollout, then multi-cancer expansion.
Validation Now
Multi-site Nottingham AI validation at 8–10 cancer centres. Pathology turnaround from two hours to under 20 minutes. CAP/AJCC-8 completeness above 95% at sign-out. Real-world data feeds the CDSCO SaMD dossier.
Commercial launch Inflection
Four regional DPaaS hub-and-spoke centres live. EHR/PACS/LIS integration at anchor sites. Adoption at three-plus paying anchor hospitals. First DPaaS revenue booked.
Scale & multi-cancer Outcome
Lung and prostate modules to production. Therapy recommendation engine across sites. MENA, SE Asia and global DPaaS launch. German MedTech OEM tie-up for screening-AI export.
Validation read-through. Whether the 94–97% accuracy and concordance figures hold across out-of-distribution cohorts at multiple sites is the single most important clinical question, and the deck flags it as the lead clinical risk, mitigated by pathologist-in-the-loop sign-out and continuous re-training.
Regulatory clock. The plan runs parallel CDSCO, FDA and EU filings, with FDA pre-submission mapped via Boston liaisons and an EU notified-body shortlist complete. Hospital procurement cycles of 9–18 months and public-tender margin pressure are named as commercial risks; the mitigants are anchor-PI equity to lock early adopters and the multi-rail revenue mix.
The device consortium. BharatScan is a consortium build — C-MET Thrissur on the probe, Suloshan Healthcare on manufacturing, DigiClinics on AI — and is at the shortlist stage on its grant tracks rather than in production. The hardware timeline is a distinct execution risk from the AI timeline.
Founder bandwidth. The deck itself names senior clinical-ML and regulatory-affairs talent as scarce and founder bandwidth as concentrated, with Series A capital tagged to senior hires in regulatory, clinical operations and business development across a Boston-Atlanta-Hyderabad structure.
The diligence list
- Multi-site validation read-out — accuracy and concordance holding across 8–10 anchor centres on out-of-distribution cohorts.
- CDSCO SaMD dossier status — filing built on real-world trial data; FDA 510(k) and EU MDR/IVDR filed in parallel.
- BharatScan consortium milestones — C-MET probe, Suloshan manufacturing; movement from shortlist to production.
- Anchor-site MoUs and PI equity terms — the lock on early adopters that the commercial plan depends on.
- Patent status on ER/PR and HER2 pipelines — the two biomarker pipelines described as patented.
- Multi-site validation read-out — whether headline accuracy holds across 8–10 anchor centres on out-of-distribution cohorts.
A sovereign-data thesis with the dataset to back it, gated on validation.
DigiClinics sits on an unusually concrete version of a familiar pitch. The problem — Western AI misreading Indian tissue while 98% of at-risk women go unscreened — is specific and well-evidenced in the deck. The moat the company leans on is the 154 TB India-trained dataset and the two patented biomarker pipelines, and the platform architecture genuinely reuses its data and imaging layers across cancer sites rather than rebuilding per disease. The policy tailwinds are real and converging.
The proposition is gated on validation. Only breast is live, at TRL 6–7; lung and prostate are earlier; the device half is a consortium at shortlist stage; and the headline accuracy figures need to hold across 8–10 sites on out-of-distribution cohorts before the rest of the platform unlocks. The thesis rests on an asset that is hard for anyone else to replicate — India-native pathology data at scale — with the open question being whether the accuracy holds under independent, multi-site validation.
