The recent development of CT technology has been remarkable, and many unique functions have been developed year by year. In this article, we first describe the requirements for diagnostic imaging and then explain the new CT technology to solve them. We introduce ultra high resolution CT as a technique to improve spatial resolution, and dual energy CT as a technique to improve contrast resolution. Model based iterative reconstruction and deep learning based reconstruction have the potential to improve both spatial and contrast resolution. After introducing the various CT techniques, we will discuss photon counting CT, which will be available in the near future.

The flow-based deep generative (FDG) models can explicitly estimate logarithm likelihood of any image. Moreover, the FDG models can generate fictional but realistic medical images from pseudo random sequences. These favorable features lead to detect anomalies in medical images, and to reconstruct three-dimensional medical images from corresponding two-dimensional medical images. To explain these applications, firstly, the discriminative model and the generative model are introduced from the viewpoint of numerical sequence. Next, the FDG models are introduced from the viewpoint of medical image processing. Lastly, two applications of the FDG models are presented : the one is anomaly detection in medical images and the other is reconstruction of corresponding chest computed tomography images from digitally reconstructed radiographs.

Accurate diagnosis of idiopathic interstitial pneumonia (IIP) is crucial as each type of IIP has a different treatment and prognosis. However, accurate diagnosis can be difficult even for specialists. In this study, we developed and tested a deep learning method for segmentation and classification of IIP using computed tomography (CT).We used 37 CT scans of patients diagnosed with different IIP types, grouped as either idiopathic pulmonary fibrosis (IPF) or non-IPF. Typical patterns of IPF in CT images are called usual interstitial pneumonia (UIP). In order to segment and classify at the same time, we labeled images with specific pixel values for cases of UIP pattern and non-UIP pattern. CT images and labeled images were used to train a 3D U-Net. The output images of 3D U-Net were used for segmentation of IIP patterns. Classification of UIP versus non-UIP was based on the pixel ratios in the output images. The results were evaluated using leave-one-out cross validation. The Dice index, which indicates the result of segmentation, was 0.774, and the overall classification accuracy was 0.757. These results show that our method may be highly accurate for the segmentation and classification of IIP using CT images.

It is difficult for radiologists to accurately classify categories of pneumoconiosis on chest radiography due to the complex pattern of lesions. Therefore, we investigated in SegNet, U-net, and Residual U-net, and developed a method for extracting pneumoconiosis region of each category with higher accuracy. In this study, a total of 54 cases of category 0 to category 3 pneumoconiosis published by the ILO etc. were used. Next, under the guidance of a radiologist, the lung field area was extracted for each category, and these were used as teacher images. Then, learning was performed using SegNet, U-net, and Residual U-net, the Jaccard index was calculated, and the accuracy of the extracted pneumoconiosis region was evaluated. In the Jaccard index in Residual U-net, category 0 was 0.98 ± 0.07, category 1 was 0.97 ± 0.04, category 2 was 0.97 ± 0.05, and category 3 was 0.97 ± 0.04. In all categories 0 to 3, the Jaccard index in Residual U-net was higher than this in SegNet and U-net, and there was a statistically significant difference (P <0.05). In the future, it will be necessary to increase the number of cases and further improve the accuracy.

Since cell becoming cancerous processes due to the accumulation of gene mutations, the effect of gene mutations other than EGFR gene might be affected imaging phenotypes of lung cancer. The purpose of this study is to clarify problems in estimating EGFR gene mutation in lung cancer by using non-invasive image examination. We collected 119 CT images and 18 RNA-Seq data from the NSCLC-Radiogenomics database and conducted experiments. The region of lung cancer was manually segmented and 365 radiomic features were determined. Linear discriminant analysis with 10 radiomic features selected by Lasso was employed for estimating the presence or absence of EGFR gene mutation. In addition, 18 RNA-Seq data with EGFR gene mutation was projected into two-dimensional space by using t-SNE. Experimental results showed that lung cancers with EGFR gene mutation have two groups with different imaging phenotypes. The patterns of gene expression level in those groups were also different. In the radiomic studies for estimating the presence or absence of EGFR gene mutation, we can conclude that it is appropriate to conduct it as a multi-group classification problem incorporated differences of gene expression pattern in lung cancer.