Lung cancer ranks 2nd among the most common cancers in Vietnam. With the ability to diagnose multiple cases simultaneously, at constant speed and accuracy and not limited by time or space, the application of Artificial Intelligence (AI) – VinDr can therefore support large-scale screening.
According to the latest statistics of the International Agency for Research on Cancer (IACR, under WHO), by 2020, Vietnam would up 7 places on the world cancer map. Specifically, our country ranks 92 out of 185 countries and territories, with a rate of 159.7 per 100,000, ranking 16th in Asia and 6 in Southeast Asia. In particular, Vietnam is one of the 50 countries with the highest cancer death rate in the world (106 / 100,000 people).
Lung cancer is one of top killers to Vietnamese people. Similar to the global general situation, in Vietnam, lung cancer, with 26,262 new cases and 23,797 deaths in 2020, was ranked 2nd among the most affected cancers in Vietnam for both genders. It is worth mentioning that up to 25% of lung cancer patients are diagnosed late, making it difficult for treatment.
Early diagnosis with various lesions
For gradually solving the lung cancer issue, the key is to increase the number of early diagnosis cases. One of the hallmarks of lung cancer is an infection that affects the respiratory tract and leads to diseases like bronchitis or other chronic infections. The chronic lung infections can be completely diagnosed early by using chest X-rays to localize the lesion. Therefore, currently chest X-ray is the first step for doctors to detect abnormalities before deeper interventions such as computed tomography (CT chest) or biopsy.
As the above fact, scientists of Vingroup Big Data Institute (VinBigdata) have put into the trial operation VinDr-ChestXR since June 2020. It is one of seven modules of VinDr – a comprehensive solution for medical image analysis that integrates Artificial Intelligence (AI) into a Picture Archiving and Communication System (PACS) to assist radiologists in making fast and precise diagnoses.
To be able to localize and classify a variety of lesions, VinDr-ChestXR is trained from more than half a million lung X-ray studies and nearly 300,000 scans in the community, especially leading Vietnamese hospitals. Collected data will then be “de-identified” and stored on the Label-PACS system for remote access and being labeled by doctors. The final result is for machine learning training. In addition, in terms of core technology, the software is also built from advanced technologies of artificial intelligence, including computer vision, deep learning, image analysis, Computer aided detection and Computer aided diagnosis.
Thus, with VinDr-ChestXR, the AI-assisted diagnostic system can detect 06 lung diseases and localize 22 common abnormalities on chest X-ray images. This is an important prerequisite step to determine the risk and progress of lung cancer in patients.
Large-scale lung cancer screening
According to statistics in 2020, Vietnam had an average of 1 doctor per 1,000 people, indicating hospital overload and enormous pressure on the medical system. Moreover, this human resource is unequally distributed among regions, leading to an increasing disparity in the quality of medical examination and treatment between rural and urban areas.
Artificial intelligence, with the collaboration of hundreds of leading doctors in the country, will become the solution to this problem. Thus, applying VinDr-ChestXR will help narrow the gap in the quality of cancer diagnosis between higher and lower level hospitals. Moreover, unlike doctors who only read each case in turn during their work time, the outstanding advantage of VinDr-ChestXR is the ability to automatically diagnose multiple cases simultaneously, working at a high speed and constant accuracy without any space or time limitation. In less than 1 second, the system was able to detect 28 common lung lesions and diseases. This is the key to reduce the overcrowding of the medical staff, thereby paving the way for a large-scale lung cancer screening.
High rate of accuracy
Implementing VinDr-ChestXR in the hospital is feasible, because artificial intelligence will not completely replace the role of the radiologist, but will provide an additional opinion for doctors to refer after reading the image. In other words, the system will be a powerful support tool, a consultant for doctors. Therefore, the application of VinDr-ChestXR means an increase in the accuracy level in disease diagnosis.
In fact, VinDr-ChestXR has been tested in major hospitals in Vietnam: 108 Hospital, Hanoi Medical University Hospital, Vimec Times City Hospital and 05 other ones in Phu Tho province. Evaluation results show that at 108 Hospital, an average of 10.5% of diagnoses changed after the doctor consulted AI, the average consensus of doctors with AI also reached 90.5%. This result was similar to that at Hanoi Medical University hospital, with the rates of 4.8% and 89.5% respectively. On average, the diagnosis accuracy of VinDr-ChestXR’s lung diseases was over 90%.
Besides VinDr-ChestXR, the system also has another feature that can diagnose lung cancer, namely VinDr-ChestCT for Chest CT interpretation. Currently, VinDr-ChestCT has been completed and will soon be put into trial implementation in hospitals. These modules promise to thoroughly and comprehensively solve the problem of early diagnosis of lung cancer for Vietnamese people.
Along with the detection of lesions and lung diseases, VinDr has many other modules, including: VinDr-BrainCT; VinDr-LiverCT; VinDr-BoneXR; VinDr-BrainMR and VinDr-Mammo. With the above features, VinDr aims to become a reliable medical imaging assistant for doctors, contributing to improving the quality of medical examination and treatment.
In addition to building VinDr solutions, from December 31, 2020 to March 31, 2021, VinBigdata organized Chest X-ray Abnormalities Detection on Kaggle, to provide a data set of 18,000 Vietnamese medical images for the domestic and foreign scientific community to develop solutions to Vietnamese health problems. See more details of the contest here.