Researchers at IIT Indore have developed a novel AI-powered solution that significantly improves early detection of various cancers by analyzing routine medical images and clinical data with high accuracy. The system combines deep learning models with explainable AI techniques to identify subtle malignant patterns often missed in conventional screening, offering a scalable, cost-effective tool for population-level cancer screening in resource-constrained settings like India.
Glimpse:
The IIT Indore AI model achieves over 94% sensitivity and specificity in detecting early-stage lung, breast, colorectal, and oral cancers from X-rays, mammograms, CT scans, and histopathological slides. Trained on diverse Indian patient datasets, it provides real-time risk scores, visual heatmaps highlighting suspicious regions, and plain-language explanations for clinicians. Already piloted in collaboration with local hospitals, the solution is now moving toward multi-centre validation and regulatory clearance, with potential to reduce late-stage diagnoses and improve survival rates nationwide.
A team of researchers at the Indian Institute of Technology Indore has developed an advanced artificial intelligence solution that promises to transform early cancer detection in India and beyond. The technology, led by Dr. Neha Singh from the Department of Biosciences and Biomedical Engineering and Dr. Abhishek Dubey from the Department of Computer Science and Engineering, integrates convolutional neural networks, attention mechanisms, and explainable AI layers to analyze medical images and clinical parameters with exceptional precision.
The AI model was trained and validated on one of the largest curated Indian cancer imaging datasets, incorporating anonymized data from government hospitals, private diagnostic centres, and collaborative institutions across multiple states. This diversity ensures robustness against variations in imaging equipment, patient demographics, ethnic diversity, and regional disease presentation patterns that often challenge globally trained models.
In rigorous testing, the system demonstrated 94–97% sensitivity and 92–95% specificity across four major cancer types: lung (from chest X-rays and low-dose CT), breast (mammograms and ultrasound), colorectal (colonoscopy images and CT colonography), and oral (clinical photographs and histopathological slides). The model not only detects malignant lesions but also stratifies risk levels, estimates tumour aggressiveness, and highlights areas of concern using intuitive heatmaps and attention overlays that clinicians can easily interpret.
What sets this solution apart is its explainability unlike many black-box AI tools, it generates human-readable reasoning for every prediction, citing specific image features (such as irregular margins, micro-calcifications, or architectural distortion) and cross-referencing with established radiological and pathological guidelines. This transparency has been critical in gaining trust from practising oncologists and radiologists during pilot deployments at partner hospitals in Madhya Pradesh and neighbouring states.
The technology is designed for real-world scalability: it runs on standard laptops and edge devices with minimal computational requirements, supports offline inference for rural screening camps, and integrates seamlessly with existing PACS and RIS systems through DICOM standards. It also complies fully with Indian data protection laws (DPDP Act) and ABDM interoperability guidelines, allowing secure linkage to patient ABHA records for longitudinal tracking.
Early pilot results from community screening camps and hospital OPDs have been promising: the AI flagged several early-stage cancers that were initially missed during routine visual reads, leading to timely biopsies and interventions. The research team, which also includes PhD scholars Priyanka Gupta and Rahul Sharma, is now collaborating with ICMR and state health departments for larger multi-centre trials and regulatory pathway support under CDSCO’s SaMD classification.
Dr. Neha Singh emphasized that the goal is not to replace radiologists but to serve as a powerful second reader that catches subtle signs in high-volume screening programs where human fatigue is a real concern. Dr. Abhishek Dubey added that the model’s ability to learn from Indian-specific data makes it particularly effective for cancers that present differently in Asian populations compared to Western cohorts.
The team has filed for patents on key algorithmic innovations and is in discussions with public health agencies and diagnostic chains for phased deployment starting later this year. If successful, this IIT Indore-developed AI could become a cornerstone of India’s national cancer screening programs and significantly contribute to reducing the country’s high cancer mortality rates through earlier detection.
“Early cancer detection saves lives and in India, it must be affordable and accessible. Our AI is built by Indians, for Indians, to catch what the eye might miss in the busiest OPDs and screening camps.”
By
HB Team
Related News
