Development of a novel artificial intelligence algorithm to detect pulmonary nodules on chest radiography.

BACKGROUND: In this study, we aimed to develop a novel artificial intelligence (AI) algorithm to support pulmonary nodule detection, which will enable physicians to efficiently interpret chest radiographs for lung cancer diagnosis. METHODS: We analyzed chest X-ray images obtained from a health examination center in Fukushima and the National Institutes of Health (NIH) Chest X-ray 14 dataset. We categorized these data into two types: type A included both Fukushima and NIH datasets, and type B included only the Fukushima dataset. We also demonstrated pulmonary nodules in the form of a heatmap display on each chest radiograph and calculated the positive probability score as an index value. RESULTS: Our novel AI algorithms had a receiver operating characteristic (ROC) area under the curve (AUC) of 0.74, a sensitivity of 0.75, and a specificity of 0.60 for the type A dataset. For the type B dataset, the respective values were 0.79, 0.72, and 0.74. The algorithms in both the type A and B datasets were superior to the accuracy of radiologists and similar to previous studies. CONCLUSIONS: The proprietary AI algorithms had a similar accuracy for interpreting chest radiographs when compared with previous studies and radiologists. Especially, we could train a high quality AI algorithm, even with our small type B data set. However, further studies are needed to improve and further validate the accuracy of our AI algorithm.

Baseline drift vector of multiple points on body surface using a near-infrared camera.

The purpose of this study was to extract the three-dimensional (3D) vector of the baseline drift baseline drift vector (BDV) of the specific points on the body surface and to demonstrate the importance of the 3D tracking of the body surface. Our system consisted of a near-infrared camera (NIC: Kinect V2) and software that recognized and tracked blue stickers as markers. We acquired 3D coordinates of 30 markers stuck on the body surface for 30 min for eight healthy volunteers and developed a simple technique to extract the BDV. The BDV on the sternum, rib, and abdomen was extracted from the measured data. BDV per min. was analyzed to estimate the frequency to exceed a given tolerance. Also, the correlation among BDVs for multiple body sites was analyzed. The longitudinal baseline drift was observed in the BDV of healthy volunteers. Among the eight volunteers, the maximum probability that the BDV per min. exceeded the tolerance of 1 mm and 2 mm was 30% and 15%, respectively. The correlation among BDVs of multiple body sites suggested a potential feasibility to distinguish the translational movement of the whole area and the respiratory movement. In conclusion, we constructed the 3D tracking system of multiple points on the body surface using a noninvasive NIC at a low cost and established the method to extract the BDV. The existence of the longitudinal baseline drift showed the importance of the 3D tracking in the body surface.

Three-dimensional shape variation and sexual dimorphism of the face, nose, and mouth of Japanese individuals.

Three-dimensional (3D) shape variations of the face and facial parts in Japanese adults were examined to collect basic data to be used for facial comparison in forensics. In total, 1000 3D facial scans (500 males, 500 females) of Japanese individuals were re-meshed into anatomically homologous shape models and analyzed by principal component analysis (PCA) after Procrustes superimposition. Facial parts (the nose and the mouth) were segmented from homologous face models and analyzed by PCA, too. Among all kinds of objects (the face, the nose, and the mouth), the most predominant shape variation represented by the first principal component (PC1) was the height-width proportion. The second largest variation (PC2) in the face and the nose was depth; for the mouth, it was the relative protrusion of the upper and lower lips. We interpreted predominant shape variations represented by the first five principal components (PCs) in each object. Asymmetric shape variations were observed within these PCs for the nose and the mouth. Sexual dimorphism of the face and the facial parts was also examined by testing the significance of sex-linked differences in PC scores. A significant difference was found between males and females for many PCs. Sexual dimorphism was examined also by emphasizing the shape difference between average male and female faces. Our results revealed predominant 3D shape variations and sexual dimorphism of the face and facial parts. The results may be informative for performing facial comparison in police investigations, an increasingly used technique.

Automatic calibration of an arbitrarily-set near-infrared camera for patient surface respiratory monitoring.

PURPOSE: A patient's respiratory monitoring is one of the key techniques in radiotherapy for a moving target. Generally, such monitoring systems are permanently set to a fixed geometry during the installation. This study aims to enable a temporary setup of such a monitoring system by developing a fast method to automatically calibrate the geometrical position by a quick measurement of calibration markers. METHODS: One calibration marker was placed on the isocenter and the other six markers were placed at positions 5-cm apart from the isocenter to the left, right, anterior, posterior, superior, and inferior directions. A near-infrared (NIR) camera (NIC) [Kinect v2 (Microsoft Corp.)] was arbitrarily set with ten different angles around the calibration phantom with a fixed tilting-down angle at approximately 45° in a linear accelerator treatment vault. The three-dimensional (3D) coordinates in the camera (Cam) coordinate system (CS; x and y are the horizontal and vertical coordinates of the image, respectively, and z is a coordinate along the NIR time-of-flight) were taken for 1 min with 30 frames per second. The data corresponding to the measurement times of 1, 3, 10, 30, and 60 s were created to mimic various measurement times. These data were used to calculate the initial matrix elements, which included six parameters of the pitching, yawing, and rolling angles; horizontal two-dimensional translation in the treatment room; and the source-to-axis distance of NIC, for a conversion from the Cam CS to the treatment room CS for which the origin was defined at the isocenter (Iso coordinate). The six parameters were then optimized to minimize the displacements of the calculated marker coordinates from the actual positions in the Iso CS. The 3D positional accuracy and angular accuracy of the conversion were evaluated. The random error of the Iso coordinates was analyzed through a relation with the angle of each measurement setup. RESULTS: Three angles of NIC and relative translation vectors were successfully calculated from the measurement data of the calibration markers. The achieved spatial and angular accuracies were 0.02 mm and 1.6°, respectively, after the optimization. Among the mimicked measurement times investigated in this study, both spatial and angular accuracies had no dependence on the measurement time. The average random error of a static marker was 0.46 mm after the optimization. CONCLUSION: We developed an automatic method to calibrate the 3D patient surface monitoring system. The procedure developed in this study enabled a quick calibration of NIC, which can be easily repeated multiple times for a frequent and quick setup of the monitoring system.