Diagnostic imaging techniques such as CT and MRI are indispensable in modern medicine.
In particular, Japan has a large number of installed imaging diagnostic devices.
Japan has the highest number of CT and MRI machines per 1 million people in the world.
Anyone can now receive diagnostic imaging.
The researchers studying a method called "radiomics" for analyzing the vast amount of information contained in these medical images are
She is Associate Professor Kamezawa Hidemi of the Department of Department of Radiological Technology Faculty of Fukuoka Medical Technology Teikyo University Fukuoka.
We are conducting research to predict the nature and prognosis of cancer from medical images, leading to better medical care.
Analyzing information based on medical images
Researching "Radiomics"
It is said that in Japan today, one in two people will develop cancer, and one in three will die from it. As the number of cancer patients increases, diagnostic and treatment techniques have improved, and the "three major cancer treatments" of Surgery, chemotherapy, and radiation therapy have been established. Currently, standard treatments are being carried out that are scientifically proven to be the safest and most effective. However, as research progresses, it has become clear that treatment outcomes vary from person to person, even for the same type of cancer.
While standard treatments determined for each type and stage of cancer are the same for many patients, personalized medicine is the treatment that is tailored to each individual patient. Cancer genomic medicine, which has been gaining popularity in recent years, has realized personalized medicine by examining the cancer genome (genes) and selecting drugs and treatments that match the genetic mutations of the cancer. However, genomic testing is quite costly and time-consuming, and there are only a limited number of facilities that can perform the testing.
One method to determine the nature of cancer is a pathology test in which cancer tissue is excised and examined under a microscope, but excising cancer is highly invasive to the body. Moreover, this method can only examine a small portion of the cancer cells. In cancers that are made up of heterogeneous cells, the characteristics of the entire cancer may be overlooked.
One promising research field is "radiomics," which is a combination of the words "radiology" and "-omics." Radiomics is a method of extracting a large number of features related to the nature of cancer from medical images and analyzing them comprehensively. It has the potential to overcome all of the challenges mentioned above, such as invasiveness, the cost and time required for testing, and the heterogeneity of cancer cells.
From CT and MRI images
Extracting features that indicate the nature of cancer
"In cancer treatment, CT and MRI scans are performed multiple times to check for the presence or absence of cancer, how far it has spread, and whether it has metastasized, but all they can see is the brightness and irregular shape of the cancer cells that are visible to the naked eye. However, medical images should contain information that reveals the properties of cancer cells, so we are trying to use this information to help with cancer treatment," explains Associate Professor Kamezawa Hidemi.
Cancer develops due to various factors, including genetic mutations and environmental factors such as smoking, but based on the hypothesis that medical images should contain the genetic mutations and anatomical/biological phenotypes (characteristics and traits) of cancer cells (tumors), the phenotypes of lesions shown in medical images are quantified and extracted as features. Radiomics is the process of analyzing these using computers to predict prognosis and recurrence.
Medical images include CT, MRI, and ultrasound (echo). Medical images are taken multiple times during the cancer treatment process and are minimally invasive. They can image the entire cancer, not just a portion of it, so they may be able to solve the problem of cancer heterogeneity in cytology. CT and MRI images also have the advantage of being able to capture the condition of the cancer and its surroundings as three-dimensional information.
It is said that there are over a hundred types of image features that can be obtained from these images. For example, digital images have pixel values (light intensity and color contrast) for each pixel, so the variation (histogram) of pixel values between adjacent pixels is used as a feature. From this, the overall shape and variation (heterogeneity) are examined. In addition, the spatial distribution (texture) of the cancer area is used as a feature, and various sizes of holes in the cancer captured in three dimensions are extracted as features and then mathematically analyzed. Another method is to decompose CT or MRI images into low and high frequency bands and extract image features specific to each frequency.
By using radiomics
Predicting head and neck cancer recurrence
If it were possible to predict the prognosis of cancer recurrence after treatment from features extracted from medical images, it could lead to better treatment options. Associate Professor Kamezawa's research is also moving forward with efforts aimed at clinical application.
Among the many types of cancer, Associate Professor Kamezawa has been conducting research on head and neck cancer, which develops in the pharynx and salivary glands from the neck to the ears. Head and neck cancer is characterized by the large number of types, including oropharyngeal cancer, parotid cancer, and nasopharyngeal cancer. The prognosis varies depending on the type, and it is considered to be a cancer area that is difficult to diagnose and treat as a whole as head and neck cancer. In particular, parotid cancer is a rare cancer with a low incidence rate, so the number of patients is small, and it is difficult to develop diagnostic and treatment techniques.
Among head and neck cancers, oropharyngeal cancer, which occurs in the middle part of the throat, is one of the causes of its development, HPV (human papillomavirus). Although not all oropharyngeal cancer patients are infected with HPV, it has been revealed that radiation therapy is more effective in patients who are infected with HPV. Therefore, Associate Professor Kamezawa has established a method to evaluate the state of HPV infection from CT images of oropharyngeal cancer using radiomics. This will help in the decision to actively select radiation therapy for patients for whom radiation therapy is effective.
In addition, for the rare cancer of the parotid gland, the malignancy is examined by a method called fine needle aspiration cytology, in which cells are extracted by inserting a needle, but this method not only places a large physical burden on the patient, but also has the problem of variable accuracy in judgment. To solve this problem, an analysis was performed to extract features from MRI images and correlate them with the level of malignancy, and it was possible to predict the malignancy with a higher accuracy (85.4%) than the conventional method (79.5%).
Recently, they have investigated and evaluated a model that extracts features without distinguishing between types of head and neck cancer and predicts recurrence. "In this research, the accuracy has only improved slightly, so we need to improve the accuracy by improving the types of features extracted and the analysis methods," explains Associate Professor Kamezawa.
Harnessing the treasure trove of medical images
Providing the best medical care for each patient
Associate Professor Kamezawa has experience as radiological technologist involved in the treatment and diagnosis of many cancer patients, and he has a strong desire to use radiomics to help with actual treatment.
One of the issues we want to solve now is the difference in quality of diagnostic images depending on the facility. University hospitals and large hospitals can introduce the latest equipment and use high-resolution images for radiomics, but not all hospitals can use the same state-of-the-art equipment. Image quality differs depending on the facility, which affects the accuracy of the analysis.
Furthermore, research is also underway into incorporating AI into radiomics. Eventually, AI will be able to automatically perform tasks from feature extraction to analysis, but for AI to intervene, even larger amounts of data will be required. In addition, it will become impossible to see how a certain result was reached. "If you can't explain why a certain prediction result was reached from that image, doctors in the clinical setting won't be convinced. The problem with AI is that it becomes a black box."
In the latest research, they are trying to pinpoint areas that are resistant to radiation therapy using medical images. Currently, radiation is irradiated to cancer cells while avoiding normal tissue, but they hope to make it more effective and minimize side effects by applying stronger radiation to resistant areas of the cancer and reducing the intensity of radiation to other areas.
"Patients have medical images taken multiple times to determine diagnosis and treatment plans, to check the effectiveness of treatment, and so on. There must be a lot of information contained in these images that has yet to be read. The goal of my research is to use radiomics to extract information from this treasure trove of medical images that will help save the lives of many patients."
The analysis results obtained through radiomics can be used as indicators to prevent recurrence and propose more effective treatment methods. Associate Professor Kamezawa emphasizes that the ultimate goal beyond that is to "cure cancer patients so that everyone can live happily." This passion is what drives the research.
