Using AI System to Diagnose Lung Cancer

Thursday - 21/05/2020 15:20

The USTH research team developed an artificial intelligence system (AI system) to read CT scan results in 5-10 seconds, while it takes doctors and medical experts 5-10 minutes to do so. 

Dr. Tran Giang Son, Project Manager - Deputy Director of Department of Information and Communication Technology at USTH, worked with his research team at USTH to develop a system that is based on data from computed tomography (CT), and trained the system to use artificial intelligence for a high-efficient calculating platform. 
 
he thong ai chan doan ung thu phoi 1 (1)
The size and location of the lung tumor automatically located by the deep learning model. (Image source: NVCC)

The team proposed the potentials of developing AI models that use nowadays most advanced technology (deep learning) to automatically detect the size and location of nodules and lung tumors on CT scans. 

Currently, the proposed system has achieved a level of accuracy at 80% on the international sample data set of more than 240,000 cases from 1,300 different studies. The data set was provided by the American National Cancer Institute (NCI), American Foundation of National Institute of Health (FNIH) and sample data collected by the research team from K Hospital. Each case includes CT scans evaluated independently by medical experts and records of diagnosed nodules and lung tumors.

With the success of the prototype, Dr. Tran Giang Son and his team aim to develop a software system to assist doctors in central hospitals and support doctors at provinces and remote areas to improve the efficiency and accuracy of lung cancer diagnosis. 

According to Dr. Tran Giang Son, the team’s long-term goal is to screen for lung cancer early and therefore improve the patients’ life expectancy. Ultimately, the team aims for an implementation of the AI system for the lung cancer screening process in hospitals and clinics across the country. 

Dr. Tran Giang Son added: “We are working on lung division techniques to exclude the areas outside the lung when reading the CT scans, which will help increase the accuracy in tumor detection and classification. We plan to continue collaborating with K Hospital and a number of provincial hospitals to build a sample data set for CT scans of lung cancer cases in Vietnam and conduct diagnostic tests on specific patients.” This is the first step in developing a prototype for supporting early screening of lung cancer throughout Vietnam. 

In Vietnam, according to the 2018 statistics, lung cancer ranked secondly in the number of both new cases (23,667 cases, accounted for 14.4% of total cases) and reported deaths (20,710 cases, accounted for 18.0% of total cases) (GLOBOCAN 2018). Lung cancer is categorized into two main types: small-cell lung cancer, accounted for 15% total, and non small-cell lung cancer, accounted for 85% total. 

The study has been published in two international journals under the SCIE Journal of Healthcare Engineering and the Journal of Real-Time Image Processing.