Cancer remains one of the leading causes of mortality worldwide, with outcomes heavily dependent on the stage at which the disease is detected. As artificial intelligence reshapes diagnostic medicine, Dr. Ankita Bansal’s research at Jio Institute explores how AI—combined with metabolic profiling—can push the boundaries of early detection from identifying existing tumours to anticipating cancer before it takes hold.

AI-enabled imaging has already demonstrated measurable improvements in screening accuracy, particularly for breast and lung cancer. Deep learning models can detect subtle radiographic features that signal early-stage disease with sensitivity often exceeding that of expert radiologists. However, imaging carries a structural limitation: by the time abnormalities become visible, the underlying biology is already well established. AI enhances detection precision, but it does not fundamentally alter what is being detected.

The more transformative progress is emerging from blood-based, multi-analyte diagnostics. Platforms that combine circulating tumour DNA, methylation signatures, and protein markers and analyse them through machine learning are beginning to identify cancer signals in asymptomatic individuals. Multi-cancer early detection (MCED) tests aim to flag multiple cancer types from a single blood draw. Although sensitivity for early-stage disease remains imperfect, the trajectory is clear: AI is enabling the extraction of meaningful signals from regimes that were previously too noisy to interpret.

What is increasingly evident is that genomic and proteomic signals alone do not capture the full picture of early malignancy. Cancer is, at its core, a disease of altered metabolism. Long before a tumour becomes radiologically visible or sheds enough DNA into the bloodstream, its cells undergo profound metabolic reprogramming, shifting how they process glucose, amino acids, and lipids to fuel rapid proliferation and survival. These metabolic changes leave detectable molecular fingerprints, particularly in the lipidome, that can serve as some of the earliest indicators of disease. Metabolic profiling, when integrated with AI-driven pattern recognition, opens a powerful new axis for early detection, one that captures the functional state of cells rather than relying solely on their genetic mutations or structural abnormalities.

What is often overlooked is that early detection is fundamentally a signal detection challenge under conditions of extreme rarity. The difficulty is not only biological but deeply computational; it requires extracting weak, heterogeneous signals from a vast background of noise. This is where AI becomes not merely helpful but essential. As algorithmic capability matures, the bottleneck shifts to data quality, cohort design, and longitudinal sampling. An emerging frontier is the development of multi-modal foundation models trained on integrated datasets spanning imaging, molecular profiles (including metabolomic and lipidomic data), and clinical histories. These models aim to move beyond single-test diagnostics toward continuous risk assessment. In this paradigm, cancer is not a binary state but a trajectory — and diagnosis becomes the identification of deviations from a patient’s baseline over time. Early evidence suggests that such approaches could detect disease signals months or even years before conventional diagnosis.

At the Jio Institute, the Cancer Metabolism and Therapeutics Lab is working at this very intersection. Our research focuses on understanding how and why metabolic reprogramming occurs in tumour cells and leveraging those insights for early detection and therapeutic intervention. We have developed a biomarker panel, identified through high-resolution metabolic profiling, that can detect stage 1 cancers with over 75% sensitivity and greater than 90% specificity. By integrating these metabolic biomarkers with AI-driven analytics and longitudinal clinical data, we are building systems designed to identify early, actionable changes before clinical symptoms emerge. The goal is not to rely on isolated biomarkers, but to detect state transitions the moment a biological system shifts from normal physiology toward disease, using metabolic, molecular, and clinical signals together.

There are, however, important reasons for caution. Many current claims about AI-enabled early detection remain insufficiently validated, particularly in diverse, real-world populations. False positives, overdiagnosis, and cost-effectiveness will ultimately determine clinical adoption. The field does not simply need greater sensitivity; it needs precision in determining who truly requires intervention and who can be safely monitored.

Nonetheless, the direction is unmistakable. Cancer diagnosis is shifting from a single, discrete event into a continuous, data-driven process. And metabolism, the engine that drives tumour biology from its earliest moments, is emerging as one of the most informative and accessible windows into that process. AI is not simply enhancing early detection; it is redefining it as the ability to anticipate disease before it becomes biologically entrenched.