- Cases of medical AI
- 2-1.Utilization example 1 HIV rapid test using a tablet terminal in rural South Africa
- 2-2. Application example 2 Detection of lesions in CT images using YOLOv3 and domain knowledge
- 2-3.Use case 3 Application of segmentation in breast tumor detection
hello! This is nishikawa from skill up ai. I am currently conducting research on image analysis and speech recognition in a laboratory related to drug discovery and medicine.
Do you know what kind of technology is used for medical AI? Recently, we often see the term “medical AI” in the news. In the news, we talk about what medical AI can do and the future prospects of medical AI, but we rarely talk about the technology used in medical AI.
Therefore, this time, we will focus on image analysis using deep learning among the technologies used for medical AI, and introduce an overview of those technologies and examples of their use.
In this section, we introduce three cases shown in Fig. 1.
Figure 1. Medical AI case study
This case study realized rapid HIV testing using tablet terminals in rural South Africa.
Pictured in Figure 2 is what is called a Rapid Diagnostic Test (RDT) kit. RDT kits are used worldwide as the most common diagnostic tool. The kit is available at a low cost, making it easily accessible to the people in rural South Africa who are the target audience for this project. However, it was very difficult to interpret the test results visually, which was a challenge for the RDT kit. Therefore, we developed a mechanism for interpreting test results using deep learning. Take a picture of the RDT kit with a tablet, input the image into a deep learning model, and let it judge whether it is negative or positive. This testing system has enabled rapid testing in rural areas where there are few doctors.
Due to the nature of the network, YOLO excels at real-time object detection. For this reason, applied research is progressing in many medical settings, such as colon examinations and CT imaging examinations. This use case uses YOLOv3 for lesion analysis on CT images. Fig. 3 shows an example output image of the constructed model.
A CT scan takes thousands of images like the one in Figure 3. It takes a lot of time even for an experienced doctor to find a suspicious lesion as a lesion from a huge amount of captured images. By supporting this work with a deep learning model, the burden on doctors can be reduced.
Early detection of breast cancer is an important issue that must be addressed globally. The most common breast cancer screening method is mammography. A mammogram is an x-ray taken specifically for the breast. An example of a captured image is shown in Fig. 4.
Determining the tumor site from the captured images is a difficult task even for specialists. Therefore, we developed a segmentation model using captured images obtained from clinical sites as learning data. A segmentation model automatically identifies the tumor site to assist the specialist in making a diagnosis.
In this blog, we introduced the technology used in medical AI. Medical AI is evolving day by day. Please pay attention to the trends of medical AI in the future.
Skill Up AI is currently offering a related course, ” AI Planning and Project Promotion Practical Course – Healthcare Edition “. In this course, you can learn the way of thinking necessary to successfully utilize AI in the healthcare field from a lecturer who is familiar with health tech. Please consider taking the course.
In addition, we hold a practical AI study session ” Skill Up AI Camp ” every Wednesday. At the study sessions, we will cover various practical themes and provide hints that will lead to improved practical skills in data analysis and AI development. There is also a corner where the instructor answers questions and concerns from the participants.