APBN New Site

APBN Developing Site

AI Enables Whole-Slide Imaging for Diagnosis of Nasal Polyps

An AI evaluation platform that uses deep learning to diagnose nasal polyps more quickly and accurately has been developed.

Nasal polyps affect more than 100 million patients worldwide and involves high management costs and poor quality of life in affected subjects. Eosinophilic nasal polyps, in particular, are linked to higher pain scores, impaired quality of life, and a high recurrence rate.

The current method to diagnose eosinophilic nasal polyps involves taking 10 random samples from thousands of polyp slides and manually counting the ratio of eosinophils to infiltrating inflammatory cells. Positive diagnosis is given when the ratio is greater than 10%. However, this leads to sampling errors and wrong diagnosis is common when the proportion of tissue eosinophils is approximately 10%.

On the other hand, whole-slide imaging takes into account the full data and thus avoids sampling bias. But it is difficult to implement in practice as it is extremely time-consuming. Using artificial intelligence (AI) for the task solves the conundrum.

Led by researchers from Sun Yat-Sen University, the team reported in the prestigious Journal of Allergy and Clinical Immunology – the top-ranked journal in allergy – development of the world’s first AI diagnostic platform for nasal polyps based on deep learning technology.

In the study, the AI diagnosed all 28 patients correctly, while the traditional method using a real pathologist resulted in 6 misdiagnoses. The AI also took a mere 5.4 minutes to perform a full film diagnosis, whereas it took a pathologist an average of 12.7 minutes to count and diagnose a patient with nasal polyps by randomly extracting 10 fields of view, and 148.6 minutes if using the full film.

By providing faster and more accurate diagnoses of nasal polyps, the authors believe their AI system “will be used widely, in particular in primary hospitals, and even all around the world through the cloud platform”. [APBN]