NCG-KCDO at India AI Summit 2026

The Challenge

There is a significant gap in clinical validation of AI models before deployment, risking inaccurate, biased, or non-generalizable outcomes in real-world oncology settings.

Most AI solutions rely on static or open datasets, lacking testing within real clinical workflows and diverse patient populations.

Absence of robust consent management systems limits patient control, regulatory compliance, and ethical data usage in healthcare AI.

The Solution

iOncology.ai is an AI-powered oncology data platform for breast and ovarian cancer diagnostics and therapeutics. The project will integrate clinical validation, a multilingual patient consent management system, and a Responsible AI framework ensuring privacy, fairness, and regulatory compliance. These modules will enable ethical, transparent, and clinically reliable AI deployment in cancer care.

Impact & Scale Potential

The platform will enable clinically validated AI models for breast and ovarian cancer, improving diagnostic accuracy and treatment decision support at scale.

Integration of dynamic multilingual consent management will strengthen patient trust, regulatory compliance, and ethical adoption across hospitals.

The Challenge

Lung cancer remains one of the leading causes of cancer mortality in India, largely due to late diagnosis. Limited access to CT scans and a shortage of radiologists exacerbate the problem, particularly in low-resource and semi-urban settings. Chest X-ray, being the most widely available imaging modality, presents an opportunity to enable early detection at scale. However, manual interpretation is constrained by delays, variability, and missed subtle findings such as early-stage nodules.

The Solution

DeepTek’s AI-powered chest X-ray solution enables early lung cancer detection by identifying nodules, masses, and 22 thoracic abnormalities with explainable outputs like heat maps and bounding boxes. Validated on 500,000+ CXRs and approved by CDSCO, CE MDR IIb, and US FDA, it integrates with PACS/RIS, works offline, and supports mobile outreach.

Impact & Scale Potential

At scale, the solution can significantly reduce diagnostic delays and missed early-stage cancers, particularly pulmonary nodules and lung masses that are often subtle on X-rays. Real-world deployments show up to 2–3x improvement in radiologist productivity and up to 60% reduction in reporting turnaround time, enabling faster referral pathways and earlier initiation of treatment.With proven deployment across 500,000+ scans and regulatory clearances (CDSCO, CE MDR, FDA), the platform is already operationally ready for national roll-out. Its seamless PACS/RIS integration, offline capability, and mobile van compatibility allow rapid expansion across NCG hospitals, district hospitals, and community screening programs.

The Challenge

Cancer care in India suffers from oncologist shortages and limited multidisciplinary expertise, especially in non-metro and government hospitals. NCG guidelines, though evidence-based, are complex and difficult to operationalize during busy clinics, leading to inconsistent adherence. Clinicians must process large volumes of unstructured patient data, resulting in delays, variability in decisions, and patient anxiety.There is no systematic mechanism to track real-world guideline concordance or ensure accountability across institutions.

The Solution

Navya Earthshot is an AI-powered decision support platform that transforms NCG guidelines into actionable, point-of-care recommendations. Using NLP and machine learning, it generates structured case summaries, patient reports, and guideline concordance tracking-reducing preparation time from hours to minutes.

Impact & Scale Potential

Navya Earthshot enables real-time, guideline-concordant decision support at the point of care, improving quality and consistency of oncology practice. By training residents and hospital staff to operate the platform, it creates a low-cost, sustainable model that does not depend on scarce specialist time. Proven deployments show faster clinical decision-making.

The Challenge

Oncology data in India is highly fragmented, siloed, and non-standardized, limiting its reuse for research and clinical decision-making.

Clinicians and researchers struggle with manual, time-consuming data discovery, spread across trials, publications, patents, and hospital systems.

The Solution

Its Trial Matching Engine analyzes patient records, treatment history, and eligibility criteria to identify suitable ongoing trials, while the Remote Phenotyping Module uses multimodal data to derive digital phenotypes for eligibility and stratification, informed by prior large-scale multimodal research. A Protocol Design & Feasibility Predictor estimates trial success probability by assessing endpoints, comparators, study duration, and investigator capacity, and an Investigator & Site Recommendation module identifies high-probability sites based on enrollment performance, experience, and publication strength.

Impact & Scale Potential

Unlike static hospital databases, Ontosight uses ontology-based knowledge graphs to unify disparate biomedical sources-trials, publications, patents, and treatment guidelines-into a semantically linked, HL7/DICOM-ready knowledge base. Traditional repositories require structured queries or IT expertise. Ontosight’s conversational AI interface allows clinicians and researchers to retrieve insights in natural language, democratizing access to advanced analytics. Beyond curation, Ontosight provides a foundation for downstream AI applications-such as trial feasibility analysis, biomarker discovery, and treatment pattern modeling-making it a strategic enabler for India’s AI-in-health mission."

The Challenge

Target volume delineation is the most error-prone and variable step in radiotherapy, with high inter-observer variability leading to frequent peer-review modifications and inconsistent treatment quality.

Existing commercial autosegmentation tools lack custom structures, transparent validation, and Indian patient representation, limiting their clinical reliability.

The Solution

DRAW is an end-to-end radiotherapy autosegmentation pipeline integrating training and inference into clinical workflows. It automates dataset curation, segmentation, validation, and performance tracking per AAPM TG263 standards. Deployed via secure, on-demand GPU VMs, it enables seamless, license-free model deployment, PHI-safe data handling, and synchronized updates across centers.

Impact & Scale Potential

DRAW enables standardised, clinically validated autosegmentation across NCG centres, reducing modification rates and improving consistency in radiotherapy planning.

Centralised deployment allows immediate access to updated models without capital investment in local GPU infrastructure, making it highly scalable nationwide.

Rapid model development supports new anatomical sites and trial-specific structures, enabling expansion beyond cervical cancer to other radiotherapy indications.

The Challenge

Cancer patients in India face fragmented care journeys, poor coordination between hospitals, and limited access to reliable information, leading to confusion, anxiety, and low treatment adherence.

Most digital solutions are hospital-centric or EMR-dependent, excluding patients in Tier-2/3 settings and low-resource environments.

Language barriers, low health literacy, and lack of continuous psychosocial support further widen inequities in patient outcomes.

Additionally, patient-generated data remains underutilized, resulting in missed opportunities for real-world evidence and research-driven care improvement.

The Solution

An integrated platform supporting cancer patients, hospitals, and researchers through three components:

• Onco App – Patient-Centric Engagement
• Symptom tracking and smart reminders for adherence.
• Multilingual AI translation of clinical instructions. Personalized education on treatment, nutrition, and self-care. Verified second opinion services.
• My Data Vault: secure, consent-based record storage and sharing.

Impact & Scale Potential

The platform improves treatment adherence, patient empowerment, and real-time clinical engagement, leading to better outcomes and reduced dropouts.

By enabling consent-driven data sharing, it accelerates real-world oncology research while maintaining privacy and trust.

Its plug-and-play, low-infrastructure design allows rapid scaling across hospitals, NGOs, and national cancer programs.

The Challenge

Current cancer diagnostics rely on siloed modalities (imaging, pathology, genomics, EHRs) that are fragmented, time-consuming, and highly subjective, leading to delayed, inconsistent, and sometimes inaccurate clinical decisions.

The Solution

The solution uses multimodal AI to integrate radiology images, pathology slides, genomic profiles, and clinical records into a single diagnostic model. CNNs and Transformers extract rich features from each data type, while fusion networks combine them for stronger predictive power. This approach captures cross-modality patterns that single-source AI misses, enabling more accurate, explainable, and scalable cancer diagnostics.

Impact & Scale Potential

By fusing imaging, genomic, and clinical data, the solution enables holistic and explainable cancer diagnosis with higher accuracy and reduced diagnostic uncertainty.It lowers dependence on scarce specialist expertise and supports standardized decision-making across hospitals.With federated learning and modular fusion models, the system can scale securely across institutions, including low-resource settings, enabling nationwide deployment.

The Challenge

Oral cancer diagnosis in India is hindered by late detection, shortage of expert pathologists, and highly subjective assessment of tumor invasion patterns, leading to inconsistent prognostication and delayed treatment decisions across care settings.

The Solution

TPhase I uses MAPSNet AI to detect fine-grained Worst Pattern of Invasion (WPOI) features in oral cancer images, and Phase II groups these patterns into three RCPath-UK categories-Cohesive, Non-cohesive, and Dispersed-usable for both biopsy and surgical specimens. This phased framework is unique in India, bringing objective AI-based pathology assessment aligned with international standards. It automates complex invasion-pattern recognition, reduces diagnostic variability, and converts detailed AI outputs into clinically actionable categories.

Impact & Scale Potential

MAPSNet enables objective and standardized WPOI classification, reducing inter-observer variability and improving clinical confidence.By translating complex patterns into RCPath-UK aligned categories, it supports actionable, guideline-based decision-making.Its compatibility with both biopsy and surgical specimens ensures seamless fit into routine workflows.With validation across multiple Indian institutions, the solution can scale nationally, strengthening oral cancer diagnostics in resource-limited centers.

The Challenge

Prostate cancer diagnosis in India is constrained by the shortage of expert uro-pathologists and the subjectivity of manual Gleason scoring, leading to inter-observer variability, delayed reporting, and inconsistent treatment decisions, especially in tier 2 and tier 3 regions

The Solution

NuGleason is an AI algorithm that detects tumor regions in prostate biopsy WSIs and assigns Gleason grades following ISUP/WHO guidelines.It segments glands into normal vs. abnormal, classifies glandular patterns, and generates a structured Gleason Score with the overall grade group.The tool improves diagnostic consistency, reducing inter-observer variability, and supports Human-in-the-Loop validation where pathologists can review or override outputs.It is interoperable with common pathology viewers and can also be deployed with PRR’s proprietary viewer (PathVu) for seamless workflow integration.

Impact & Scale Potential

NuGleason enables standardized Gleason scoring, reducing diagnostic discordance and improving clinical decision-making.Its human-in-the-loop design supports pathologists by automating routine cases while preserving expert oversight for complex ones.Being interoperable with existing pathology viewers, it allows seamless adoption without disrupting workflows.With cloud-native deployment, the solution can scale across NCG hospitals

The Challenge

India faces a severe lung cancer burden driven by late-stage diagnosis, acute shortages of radiologists, fragmented imaging infrastructure, and limited access to screening in rural and underserved regions. These systemic gaps result in high misdiagnosis rates, long reporting times, and mortality exceeding 80% in low-resource settings.

The Solution

DecXpert is an AI-driven lung cancer screening platform that detects and characterizes nodules with over 95% accuracy in under a minute. Built for India’s cancer care needs, it reduces misdiagnosis 4x, improves early detection by 50%, and is 4x more affordable. Its adaptive, CDSCO-compliant architecture integrates easily into NCG hospital workflows, supporting scalable and equitable screening.

Impact & Scale Potential

DecXpert enables rapid, high-accuracy early detection significantly reducing diagnostic delays and missed cases.Its edge-based, online/offline design allows deployment even in low-connectivity environments across NCG hospitals.Being more affordable and vendor-agnostic, it supports large-scale population screening at minimal incremental cost.

The Challenge

Breast cancer in India is often detected at advanced stages due to limited screening infrastructure, high patient load on radiologists, and reliance on single-modality imaging, leading to diagnostic delays and variability. The lack of integrated, explainable AI tools further constrains consistent and accurate interpretation across diverse clinical settings.

The Solution

An AI-driven diagnostic platform that integrates mammography, ultrasound, and PET/CT data to enhance early breast cancer detection and risk stratification. Using CNNs, ResNet/DenseNet models, and U-Net segmentation, it delivers automated, explainable lesion analysis to support radiologists and reduce diagnostic variability. Validated on retrospective datasets and aligned with Indian clinical needs, it is at TRL 5–6 with IEC approval and ready for NCG pilot deployment.

Impact & Scale Potential

The solution enables earlier and more accurate detection through multimodal AI, reducing missed diagnoses and improving patient outcomes.By functioning as a decision support system, it enhances radiologist efficiency and standardizes care across high-volume centres.Its cloud-enabled, modular architecture allows rapid scaling across NCG hospitals and integration with existing radiology workflows

The Challenge

Cervical cancer screening in India remains critically low (~1.9%) due to high costs, lab dependency, and limited rural access.

A shortage of specialists and urban concentration of services leads to diagnostic delays and missed early-stage cases.

Women face geographic, economic, and socio-cultural barriers, including long travel distances, stigma, and low health literacy.

Frontline health workers lack simple, real-time, decision-support tools, resulting in late detection and preventable mortality.

The Solution

AQH and Nitara Health’s CerviScreen uses AI-powered mobile cervicography to improve early detection, diagnostic accuracy, and efficiency. By addressing systemic challenges in public health, it supports to deliver equitable, patient-centric screening in underserved regions, bridging the gap to advanced technologies.

Impact & Scale Potential

The solution enables early detection at the last mile, significantly reducing late-stage diagnoses and preventable cervical cancer deaths.

By lowering cost and eliminating lab dependency, it supports population-scale screening across PHCs, outreach camps, and community programs.

Its AI-driven, mobile-first architecture allows rapid replication across states and integration into national programmes like NCG.

With continuous learning from real-world data, it has strong potential for nationwide and global deployment in low-resource settings.

The Challenge

Cervical cancer screening in India remains critically low ( < 2% ) due to high costs, invasive procedures, long turnaround times, and limited laboratory access. Existing methods provide delayed and binary results, offering little insight into disease progression or urgency of intervention.Rural and underserved populations face severe access barriers, including stigma, lack of specialists, and poor infrastructure.The absence of rapid, affordable, and point-of-care diagnostics leads to late-stage detection and preventable mortality.

The Solution

An AI-powered biosensor for rapid cervical cancer screening delivers results in 5–10 seconds, detecting disease-specific molecular signatures with high sensitivity. Beyond HPV detection, it provides stage-specific prognosis using self-learning AI trained on large clinical datasets. The device offers actionable insights for immediate referral, works in real time at the point of care, and is affordable and non-invasive

Impact & Scale Potential

This solution enables ultra-fast, affordable cervical cancer screening at the point of care, dramatically increasing early detection in low-resource settings.By delivering stage-specific prognosis within seconds, it supports timely referral and reduces late-stage diagnosis and mortality.Its non-invasive, portable, and low-cost design allows large-scale deployment across PHCs, rural clinics, and community programs.

The Challenge

Screening approaches like HPV, Pap smear and colposcopy are resource-heavy and unavailable at PHCs. While VIA is cost-effective in low resource settings, its accuracy is dependent on the examiner’s experience, leading to inaccurate results. Experts are not available in rural settings, available health workers are not skilled enough. Second opinion is not feasible since cervix appearance is not documented. Some solutions were proposed for task shifting using AI. However, it requires internet since it uses cloud based AI services. To address this challenge, SAKHI Manipal VIA-AI prototype was developed—an AI-enabled device designed to support healthcare professionals (HCPs) in conducting cervical cancer screening.

The Solution

An offline-capable AI tool for cervical cancer screening deployable by public health workers in low-connectivity areas.It uses explainable AI to highlight abnormal regions in cervical images and compares performance against benchmarks.The project includes device refinement for commercialization with improved imaging, user guidance, and instant AI-based reporting.

Impact & Scale Potential

This solution enables early, accurate cervical cancer screening at the community level, empowering frontline health workers in low-resource settings.Its offline, portable and explainable AI design ensures reliable use across rural and remote areas without specialist dependency.Built-in documentation and referrals strengthen continuity of care, second opinions, and workforce training.Once validated, the device can be scaled nationally across PHCs and adapted for other LMICs, creating a sustainable global model for AI-enabled screening.

The Challenge

Cancer diagnosis in India is hampered by late detection, fragmented diagnostic workflows, and limited access to specialized expertise, especially in Tier-2/3 cities and rural regions. Radiology, pathology, and clinical data are typically analyzed in isolation, leading to delays, inconsistent interpretations, and missed early-stage cancers. The shortage of oncologists and advanced diagnostic infrastructure further increases diagnostic burden, resulting in high variability in outcomes and preventable mortality.

The Solution

CT images of breast cancer patients are given to the pretrained CNN and Vision Transformer model for segmentation of the cancer-affected portion and classification of the level of cancer. Based on the classified level, the immunotherapy required is provided with the support of Predictive AI. Dose distribution and optimisation of radiation delivery are done entirely by the transformer-based network, which is trained using retrospective data of breast cancer patients and can be further clinically validated with prospective data and the clinical team involved in the project. This radiation planning and therapy prediction will reduce the cost of treatment, as contouring and planning are automated and useful in resource-limited settings, especially in rural India.

Impact & Scale Potential

The platform enables early, accurate, and explainable cancer screening by unifying multimodal clinical data into a single AI-driven decision-support system, reducing diagnostic turnaround time by up to 40%. Its cloud-to-edge architecture allows seamless deployment across primary health centers, district hospitals, and tertiary care institutions without requiring major infrastructure upgrades. This makes it highly scalable for national screening programs, supporting population-level cancer surveillance, reducing specialist workload, and ensuring equitable access to high-quality diagnostics across diverse healthcare settings.

The Challenge

Current chemotherapy dosing is primarily based on Body Mass Index (BMI), which fails to reflect true body composition and does not account for sarcopenia-a critical factor influencing drug toxicity and treatment tolerance. This leads to widespread over- or under-dosing, higher adverse events, treatment interruptions, and inconsistent outcomes, especially in Indian patients where body composition differs significantly from Western reference populations. There is no standardized, population-specific biomarker available to guide personalized chemotherapy dosing at scale.

The Solution

The solution leverages skeletal muscle mass index (SKM) as a superior metric to BMI for personalizing chemotherapy dosing. We publish reference SKM ranges derived from CT scans of the Indian population, linked to prescribed chemo doses and patient outcomes. Using automated skeletal muscle segmentation on CT images, we assess sarcopenia and guide chemo dose adjustments. The segmentation algorithm, trained on a combination of public and private datasets, uniquely employs semi- and unsupervised learning leveraging state-of-the-art models like SAM, CLIP, and vision-language embeddings. This approach allows scalable learning from 20 years of retrospective cancer data at Tata Memorial Hospital, overcoming the limitations of purely supervised methods.

Impact & Scale Potential

This solution introduces Skeletal Muscle Mass (SKM) as biomarker for chemotherapy personalization, enabling precise dose optimization using routinely acquired CT scans. By automating SKM extraction through AI and linking it to 20 years of real-world dose-response data, the platform can be deployed across oncology centers without additional tests or infrastructure. Its semi- and unsupervised learning framework allows continuous improvement across large, diverse datasets, making it highly scalable for national cancer networks, reducing toxicity, improving treatment adherence, and establishing India as a global reference for evidence-based, AI-driven chemotherapy dosing.

The Challenge

In oncology care, critical symptoms and medication issues often go unreported between clinical visits, especially due to language barriers, low digital literacy, and patient hesitation. This creates a visibility gap for clinicians, leading to delayed interventions, preventable complications, and increased emergency visits. Clinics are forced to rely on fragmented phone calls and messages, increasing staff workload while still missing early warning signs.

The Solution

Meeval enables patients (or caregivers) to log symptoms and meds in English or Malayalam, get gentle nudges, and share clear timelines with their care team. A simple rules-based triage highlights potential red flags for clinician review. The WhatsApp conversational layer and advanced analytics are planned, not live.

Impact & Scale Potential

Meeval enables real-time, language-inclusive symptom and medication tracking, giving clinicians structured visibility into patient status between visits. By improving reporting adherence and accelerating time-to-review for red-flag symptoms, it reduces avoidable complications and clinic message burden. Its lightweight, EMR-agnostic design allows rapid deployment across oncology centres, and the pilot-generated workflows and metrics create a replicable playbook for national scale—making continuous patient monitoring affordable, inclusive, and operationally sustainable for Indian cancer care systems.

The Challenge

Hospitals generate vast volumes of heterogeneous data across EHRs, labs, imaging, genomics, and clinical notes, but this data remains fragmented, unstructured, and underutilized for real-time clinical decision-making. Clinicians face information overload, limited cross-modal insights, and increasing risk of diagnostic errors and adverse drug events. As a result, critical patterns are missed, preventable complications occur, and valuable clinical data is rarely converted into actionable or publishable intelligence.

The Solution

A multimodal AI engine integrating structured/unstructured hospital data (EHR, labs, imaging, genomics, lifestyle). Provides real-time reasoning, prescription checks, predictive risk scoring, Molecular tumor board, Multi specilaty board and research analytics.

Impact & Scale Potential

The platform significantly enhances clinical safety by preventing adverse drug events, enabling early detection of complications, and supporting faster, more accurate diagnoses. By converting routine hospital data into structured, analyzable outputs, it also accelerates clinical research and evidence generation. The agentic architecture-using specialized AI agents such as OncoAgent, CardioAgent, and OrthoAgent-allows rapid customization across departments without rebuilding core systems. The solution demonstrates strong readiness for scale, with the potential to become a foundational AI layer for hospital intelligence globally.