AI-powered blood tests spots rare eye cancer earlier than standard methods

A major breakthrough has been made that a simple blood test analyzed by artificial intelligence could allow for early detection of primary vitreoretinal lymphoma (PVRL), a rare and aggressive cancer often considered misleading for inflammation.

A learning model, designed to analyze patterns, was trained by researchers using routine complete blood count data from 255 PVRL patients and 292 controls. This technique is a noninvasive diagnostic technique that aims to replace the currently dominant invasive position.

The six-feature random forest model attained an area under the curve (AUC) of 0.85 in the discovery cohort, with continuous verification across groups at AUC 0.80-0.83.

This result surpasses conventional biomarkers, the interleukin-10/interleukin-6 ratio, which only scored 0.65-0.78.

PVRL commonly presents as uveitis with blurred vision, hazy sight, and diagnosis is often deferred by months due to unspecified symptoms.

The model specifically identified 38 PVRL cases among 66 high-risk individuals and 2 more among 83,610 low-risk patients.

The data presented 95.0% sensitivity, 99.97 specificity, 57.6% positive predictive value, and 99.9% negative predictive value.

The recent study underscores PVRL’s fatality when it is metastasized to the brain, often bilateral and affecting older adults with less pain.

Nonetheless, a free web application allows therapists to input blood results from instant risk scores, seeking immediate care for vision loss.

In addition, this blood-based strategy could contribute to save vision and lives, circumventing expensive imaging or lumbar punctures while experts called it a significant discovery for this misdiagnosed malignancy.