AI in Healthcare: From Clinical Tool to first line of defense

During one of the executive team meetings at an AI startup focused on biopharma, I posited that they should think of creating a version of their prognostic AI models that can be pushed to the edge, as-in edge models. These models could then be integrated in wet-labs and into the devices that become your first line of defense in an overstretched healthcare system.

There were muted scoffs and suppressed smiles, but I pushed ahead. Today the healthcare diagnosis is done by reading scans, detecting abnormalities, analyzing reports, and then supporting the clinical decision based on current standard-of-care and known-knowns. Generative AI can (figuratively) think outside the box. I left disappointed as no action items were taken in that meeting, although I wished I could get everyone to at least agree on a proof-of-concept.

If the lab analyzers, like immunoassay, hematology, or biochemistry, are able to generate a prognostic report along with the diagnostics report, the doctors will be able to help their patients prevent diseases rather than treat the symptoms after the fact. Integrate some of these models in your smart watches, OTC (over the counter) devices, home, and you push the health awareness where it belongs, to the individual. The possibilities are endless and the most exciting opportunity is what I would call ‘digital organ’ on the lines of organ-on-a-chip (OoC).

Imagine an AI model that predicts how a heart’s rhythm would change under stress or medication. The researchers could combine physiological equations with AI trained on experimental and clinical measurements, validate its predictions against independent biological data, and deploy a compact version on a low-power chip. Sensors could supply real-time inputs, while the model produces predictions and other data.

This could lead to the holy grail of personalized and precision medicine by using a patient’s scans, lab results, medical history, and sensor readings that could turn the model into a digital twin of selected organ functions. Clinicians could then explore how that person might respond to different treatments or doses. Stretch it a bit further, and the lab doing a biopsy on your tissue can also synthesize a drug that is customized to your specific cancer!