|
Identification of Nodular Shadows from CT Images Using Improved CoAtNet Incorporated Clinical Recording |
|
Yuto Nishitaki, Tohru Kamiya(Kyushu Institute of Technology, Japan), Shoji Kido(Osaka University, Japan) |
| The number of images obtained in a single chest CT scan is enormous, placing a heavy burden on the physician who reads the images. Therefore, computer-aided diagnosis systems have been introduced to reduce the burden on the physician and reduce the number of undetected lesions. In this paper, we propose a nodular shade identification model using deep learning, aiming to improve diagnostic accuracy by introducing medical record information in addition to image information. Experimental results show that the accuracy of nodal shade discrimination improves when medical record information is added. |
|
|
Classification of Driver Gene Mutations from 3D-CT Images Based on Radiomics Features |
|
Shion Watanabe, Tohru Kamiya(Kyushu Institute of Technology, Japan) |
| Lung cancer is the most common cause of death from cancer. Diagnosis is made mainly by biopsy, which checks for mutations in the driver genes of lung cancer. If there is a mutation, molecularly targeted drugs with higher therapeutic efficacy can be used. However, it is difficult for physicians to make a decision and places a heavy burden on patients. To solve this problem, a computer aided diagnosis (CAD) system is needed to identify the presence or absence of driver gene mutations from CT images. In this paper, Radiomics is used to extract features from 3D lung cancer in CT images and analyze them using machine learning to classify the presence or absence of mutations. |
|
|
An Denoising Method Low-Dose CT Image Using Image Restoration Convolutional Neural Network |
|
Yuta Sadamatsu(Kyushu Institute of Technology, Japan), Seiichi Murakami(Junshin Gakuen University, Japan), Guangxu Li(Tiangong University, China), Tohru Kamiya(Kyushu Institute of Technology, Japan) |
| Radiation exposure from Computed Tomography (CT) scans may pose a future risk of cancer. Efforts are therefore being made to reduce radiation exposure. During the examination, noise is generated in the image when the dose is reduced. In this study, we focus on Convolutional Neural Network (CNN), a type of deep learning model that has demonstrated high accuracy in image processing tasks. We aim to achieve higher accuracy by employing a channel attention module and a loss function MAE. Using whole body slice CT images of pigs, we evaluate the image quality by Peak Signal-to-Noise Ratio (PSNR) and show that the proposed method is effective. |
|
|
A Detection Method for Nodular Shadows from Temporal Subtraction Images Using A Machine Learning Technique Incorporated Radiomics Features |
|
Natsuho Baba, Tohru Kamiya(Kyushu Institute of Technology, Japan), Takashi Terasawa, Takatoshi Aoki(University of Occupational and Environmental Health, Japan), Shoji Kido(Osaka University, Japan) |
| In this paper, we focuses on the automatic extraction of lung cancer and attempts to develop a computer-aided system. One of the CAD techniques is temporal subtraction, which is a subtraction operation using current and past images of the same patient to remove normal structures and emphasize newly appearing areas that have changed over time. However, because two images taken at different dates are used, artifacts due to misalignment often occur. Therefore, using the temporal subtraction image generation method, candidate regions of abnormal shadows are extracted as initial shadows, and binary classification is performed to determine whether they are lung cancer or not. |
|