This AI Urine Test Detects 50 Diseases and Reads What Genetic Tests Miss | #193

What if a single urine sample could screen for 50 diseases before you ever felt sick? Dejan Nenov, co-founder of AI diagnostics company Luventix, joins Dr. Joy Kong to explain how.

Dejan Nenov has spent nearly a decade navigating FDA processes and clinical trials, and also founded Panaton, a healthcare IT company connecting hospitals, labs, and clinics across the US. Luventix uses machine learning to analyze 38,250 raw chemical data points in a single urine sample, identifying disease patterns without isolating a specific biomarker first. Its ongoing IRB-approved study covers roughly 1,250 patients across colorectal cancer, Crohn's, celiac disease, and SIBO. The same model can predict which patients will respond to a given treatment, a tool called companion diagnostics, and could eventually screen for dozens to hundreds of conditions from one low-cost sample. Dejan and Dr. Joy also cover the regulatory pathway for AI diagnostics, including the FDA's Lab Developed Test framework, and how genetics and metabolomics offer two different, complementary pictures of health.

The conversation also covers a personal case: a close friend's daughter whose rare condition took years and specialists across ten states to diagnose, and why Dejan believes AI could have caught it sooner.

Dr. Joy and Dejan close with what a fully AI-run hospital already looks like in China, and Dejan's vision for an at-home test, like a pregnancy test, that catches disease three to eighteen months before symptoms start.

Dejan talks about:

00:00 How a dog's nose inspired Luventix

03:58 A dog's nose as a gas chromatograph

04:22 Training AI to classify disease

05:25 The 450-patient GI cancer trial

06:54 Urine reflects the body's metabolic state

08:07 Skipping biomarkers, reading raw data patterns

11:00 38,250 data points per urine sample