Let AI take a look at the X-rays taken in the hospital!

Import:

In daily life, people inevitably have to go to the hospital for checkups. But did you know that AI has quietly entered the medical diagnosis process and participated in the battle to protect human health?

In this battle of defense, AI has performed particularly well in medical image recognition. Let’s start with AI image recognition and see how AI+medical care can help society realize its technological achievements.

Difficulties in medical image recognition:

When we go to the hospital for medical treatment or physical examination, we may get a large X-ray, CT or MRI film, but it is not easy to interpret these medical images efficiently and accurately.

When taking CT and MRI images, it is not just one picture, but dozens or hundreds of pictures are scanned at the same time. The radiologists will then select the most likely lesions from so many pictures and give them to other doctors for reference, which is extremely labor-intensive.

According to the 2018 White Paper on Medical AI Technology and Application, my country's medical imaging data grows by 30% each year, while the number of medical imaging doctors only grows by 4.1% each year. It is not uncommon for an imaging doctor in a tertiary hospital to look at tens of thousands of medical images every day.

Under such a high workload, even well-trained doctors are bound to make mistakes.

Even with medical imaging, some diseases are difficult to diagnose in the early stages. Let’s take lung cancer as an example. According to the 2022 data from the International Agency for Research on Cancer (IARC), lung cancer is the cancer with the highest incidence and mortality in China.

According to the "Chinese Expert Consensus on the Diagnosis of Early Lung Cancer (2023 Edition)", if lung cancer can be diagnosed in the early stages (stage I) and actively treated, the five-year survival rate can reach 77%-92%. However, if it is diagnosed in stages III to IV, the five-year survival rate is only 10% to 36%.

However, stage I lung cancer has no obvious clinical symptoms and is difficult to identify from X-rays. Low-dose spiral CT scans are required, which requires radiologists to carefully distinguish a large number of CT images. In the early stages of lung cancer, some lesions are mixed with benign nodules and are difficult to distinguish.

AI can do better:

After the emergence of convolutional neural networks (CNN), AI has made rapid progress in the field of image recognition. At the same time, with the help of deep learning algorithms, AI can use labeled medical images for learning, thereby assisting human doctors in making diagnoses. In the diagnosis of certain diseases, their performance is even comparable to that of top human experts.

For example, a 2019 study demonstrated the advantages of AI in the early diagnosis of lung cancer.

The researchers used more than 45,000 annotated chest CT scans to train the algorithm. After training, the scientists asked the AI ​​model to analyze new CT images and compared the results with those given by six professional radiologists.

The results showed that the AI ​​missed diagnosis rate was 5% lower than that of human doctors. The AI ​​false positive rate was also 11% lower than that of human doctors. False positive can be understood as the situation where there is no lung cancer but it is misdiagnosed as lung cancer. This shows that the detection accuracy of this AI model is no less than that of professional human doctors.

If such a model can become popular, it will undoubtedly greatly alleviate the workload of doctors and allow more people to receive treatment as early as possible.

And the popularization of such technology has already begun.

In 2021, some of our country's top three hospitals have introduced AI-assisted diagnosis systems for small lung nodules. This system can distinguish whether nodules in CT images are lung cancer lesions, and it can identify small nodules that cannot be distinguished by the naked eye, thereby assisting doctors to make more accurate judgments.

It is worth mentioning that in January 2024, this system was also piloted for the first time in two grassroots hospitals, Xinhua Hospital in Tongzhou District, Beijing, and Yongledian Community Health Service Center. If such a system can be popularized in grassroots hospitals, it will greatly alleviate the problem of unbalanced medical resources.

Of course, we only cited the example of early diagnosis of lung cancer. AI can do much more than that. AI can play an important role in the diagnosis of breast cancer, examination of diabetic retinopathy, photo recognition of skin diseases, diagnosis and screening of cardiovascular and cerebrovascular diseases, etc.

The emergence of AI-assisted diagnosis may greatly change the medical and health industry and safeguard people's health.

AI image recognition still has shortcomings:

It is undeniable that the accuracy of AI recognition depends on high-quality labeled data. However, the labeling of medical imaging data is very different from that of general data.

For example, ordinary people can annotate the training data set for self-driving cars, because ordinary people can easily distinguish objects such as roads, traffic lights, pedestrians, bicycles, etc. in pictures.

However, the labeling of medical imaging data relies on experienced doctors, which makes it more difficult to obtain training data.

However, medical image recognition and analysis is a relatively new research field. A large number of scientists around the world are engaged in research in this field. Associations such as MICCAI have also integrated a large number of experts in computer science, medicine, engineering and physics, and produce thousands of academic papers every year.

I believe that in the future, AI technology will continue to promote the development of the medical and health industry, allowing more people to enjoy better medical care.

References:

Han B, Zheng R, Zeng H, et al. Cancer incidence and mortality in China, 2022[J]. Journal of the National Cancer Center, 2024, 4(1): 47-53.

Chinese Society of Respiratory Medicine. (2023). Chinese expert consensus on the diagnosis of early lung cancer (2023 edition). Chinese Journal of Tuberculosis and Respiratory Diseases, 46(1), 1-18.

Liu Yunqin, Li Shengjin. Research progress in early diagnosis of lung cancer[J]. Advances in Clinical Medicine, 2024, 14: 2406.

Ardila D, Kiraly AP, Bharadwaj S, et al. End-to-end lung cancer screening with three-dimensional deep learning on low-dose chest computed tomography[J]. Nature medicine, 2019, 25(6): 954-961.

2018 Medical AI Technology and Application White Paper

Author: Yunjiyu Science Creation Team

Reviewer: Qin Zengchang, Associate Professor, School of Automation Science and Electrical Engineering, Beihang University

The article is produced by Science Popularization China-Creation Cultivation Program. Please indicate the source when reprinting.