Building a large data warehouse
Speaking with Lao Dong reporter, Prof. Dr. Nguyen Huu Sau - Deputy Director of the Central Dermatology Hospital, senior lecturer of the Department of Dermatology - Hanoi Medical University, said that the State-level research topic on AI has entered the acceptance phase with positive results.
According to Prof. Dr. Sau, the idea of applying AI in tooth decay has been around since 2008 when he had just completed his study program in France. Here, hospitals have used teledermatology - a remote diagnosis based on skin damage photography - for many years.
Another reality that prompted the research team to get to the topic was the shortage of dermatologists in provinces and districts (old).
In 2017, he met Associate Professor, Dr. Nguyen Long Giang - Deputy Director of the Institute of Information Technology - Vietnam Academy of Science and Technology.From here, the two sides started to develop a State-level research topic on the application of AI in the diagnosis of three diseases: psoriasis, atopic dermatitis and skin cancer.
To build a reliable AI model, the research team needed a huge data warehouse. During the implementation process, doctors at the Central Dermatology Hospital, 108 Central Central General Hospital and Thanh Hoa Dermatology Hospital participated in taking photos, classifying and labeling each lesion.
We collect from 9,000 to 12,000 photos of each disease. A patient can have multiple scans, at many different stages. We also took more similar diseases to help AI learn to distinguish, said Prof. Dr. Sau.
All photos are consulted by experts to ensure the highest accuracy before being put into the system.
After many years of testing, the research team has built 3 applications: software (app) for patients, allowing them to take photos of skin damage and receive suggestions in 20 - 30 seconds; App for doctors, supporting diagnostic orientation; and a system used in research and image management.
The test results at the 3 hospitals are all very positive. The sensitivity and specificity in diagnosis are all over 85% - a level that even specialists often find difficult to achieve in practice, especially in diseases with unusual symptoms or when the patient has been Omed.
Prof. Dr. Sau said that an internal experiment was conducted, a group of doctors diagnosed on the lesion photo and then the AI also diagnosed on the same photo set.
The result is AI for higher accuracy. Not all doctors are deeply trained, and many cases of transformed lesions make doctors confused, Mr. Sau explained.
AI cannot replace doctors, and it is impossible to prescribe
However, although AI gives high diagnostic results, Prof. Dr. Sau affirmed a consistent view: AI does not replace doctors.
The final decision maker is always the doctor. AI is just a support tool to help diagnose faster and more accurately, especially for lower-level doctors or patients who cannot see a doctor immediately".
However, the AI application has not yet been officially deployed at the Central Dermatology Hospital because the topic is in the acceptance stage. However, when asked about the expansion roadmap, Prof. Dr. Sau said that the hospital and industry leaders are very supportive, especially in the context of Resolutions 72 and 57 strongly promoting digital transformation in healthcare.
In the future, the research team hopes to expand the software to cover more dermatological diseases, even integrating initial consultation for patients after receiving AI results, but not the prescription or treatment stage.
Prof. Dr. Sau said that if the application is successful at the Central Dermatology Hospital, the group will evaluate the effectiveness and propose widespread implementation.
