One of the applications included in the jointly researched image reading support system is " Visualization of Temporal Changes in Bone Regions", which is intended to visualize temporal changes in bones in comparison between current and prior images.

Because loss of mobility due to bone metastases can have significant effects on daily activities, early detection and treatment are essential to help maintain the patient's QOL. But bone metastases are often asymptomatic in the early stages, so early detection of bone metastasis in routine follow-up CT exams is expected.

In such follow-up CT exams, priority is given to carefully checking the recurrence of cancer and organs/lymph nodes where metastases are most likely to occur, but bone metastases are considered very difficult to be detected. Moreover, there are numerous bones in all parts of the body, so a large number of CT images must be read in whole-body exams. To address these challenges, Kyoto University and Canon have been working together to develop a system that can help physicians evaluate changes in bone over time in CT images. The basic approach for the detection of metastases is to compare current CT images and prior CT images in order to identify any differences between them. Subtle differences between the images are clearly visualized by generating subtraction images to assist in the detection of bone metastases.

Dr. Sakamoto states, “I feel we should consider CT exams an opportunity to detect abnormalities such as bone metastases in the early stages, given that CT is a commonly used imaging modality. We shouldn’t miss even the smallest bone lesion, because bone metastases can progress and lead to bone destruction, making it difficult for patients to continue their daily activities as members of society."

The key to generate useful subtraction images is to ensure accurate spatial registration between current and prior images. In CT exams, the body shape and patient position on couch can change in different exams, and even a way of breathing can affect the acquired images.

To overcome this problem, Dr. Sakamoto suggested employing a technique that is already used for the analysis of clinical images and satellite images. This technique can also be used to correct misregistration of target structures with varying shapes in CT images by employing a method in which the shapes in the images are warped or deformed in order to achieve accurate registration. First, the image is deformed to roughly align the body structures that are clearly different in current and prior images, such as hip height and angle. The images are then categorized into torso images, which are less likely to be affected by body position, and non-torso images (such as images of the extremities), which are more likely to be affected by body position. Each image is then deformed to achieve precise registration. Finally, the images are integrated, and residual noise is removed to obtain subtraction images that clearly show the changes in bone, even if the images were acquired under different conditions. However, for this method to be practical for use in the clinical setting, it was considered essential to increase the speed of image processing and the accuracy of image registration.

By developing original algorithms and improved image processing techniques, Kyoto University and Canon have introduced new image registration technology for generating subtraction images specifically intended to display changes in bone over time, even if the images were acquired using different scanning conditions. In addition, an advanced labeling function has been implemented to indicate the locations of the vertebrae and ribs in the subtraction images. This makes it much easier to identify the locations of lesions. In fact, it has been reported that the system can reduce the image reading time by two-thirds in some cases.

About a year has passed since the system entered clinical operation at Kyoto University. Dr. Sakamoto notes, “It's my impression that the detection of bone metastases is much easier than it used to be. Other physicians have commented that the subtraction images allow them to see at a glance that there are no bone metastases, which gives them a greater sense of security. We feel that the system not only shortens image reading times, but also helps to reduce the mental stress on physicians."

The lung is a region for the frequent development of cancer metastasis. And because the lung plays a central role in respiration, lung metastases can have a significant impact on the patient's QOL. However, lung nodules (small nodular shadows observed in lung tissues) suspected to be metastases are considered difficult to be detected when they are small. And even large lesions may be difficult to identify as lung nodules in some areas. Nevertheless, the larger the nodule, the more likely it is to be cancer, so it is important to observe changes in nodules over time. In actual image reading, current and prior images are placed side by side and compared one at a time. Because the shape of the lungs varies due to respiratory motion, it is necessary to adjust the position of the images while performing comparison. This not only requires time and effort but also causes mental stress due to the possibility of missing lesions.

Kyoto University and Canon have therefore worked to conduct research in a system that supports not only image reading for the bones, but also for the lungs. This work has led to the introduction of a new ROI (region of interest) comparison function. The ROI comparison function combines Canon's unique advanced image registration technology with existing technologies to extract ROIs containing characteristic area in lung tissue from chest CT scan data and measure their size and volume using AI-based analysis and processing. Current and prior images are displayed side by side, and the extracted ROIs in the lungs are linked and tracked over time, allowing any changes in size or other characteristics to be seen at a glance.

Even faster real-time performance must be achieved in order for image reading support systems to gain widespread acceptance in clinical situations where every second counts.

Dr. Sakamoto states, “We want to continue exploring new ideas together with Canon to further improve the system. The image reading support system allows abnormalities to be detected in the early stages, which means patients promptly select appropriate treatment. This leads to a better prognosis and an improved QOL for our patients."

Canon will continue working together with Kyoto University to further refine its technologies while responding to the needs of clinicians on the frontlines of healthcare. The ultimate goal is to create a society in which cancer patients can continue to enjoy their lives as they wish.

• *AI is used only during the design stage. The Image Reading Support System itself does not have self-learning capabilities.
• *The software does not diagnose bone metastases or pulmonary nodules.
• *Clinical comments are based on the judgment of a physician.

Recommended Articles

Read articles in ”Realizing Professionals' Ideals”