Background: Chest X-ray (CXR) imaging is the most common examination; however, no automatic quality assurance (QA) system using deep learning (DL) has been established for CXR. This study aimed to construct a DL-based QA system and assess its usefulness. Method: Datasets were created using over 23,000 images from Chest-14 and clinical images. The QA system consisted of three classification models and one regression model. The classification method was used for the correction of image orientation, left–right reversal, and estimating the patient’s position, such as standing, sitting, and lying. The regression method was used for the correction of the image angle. ResNet-50, VGG-16, and the original convolutional neural network (CNN) were compared under five cross-fold evaluations. The overall accuracy of the QA system was tested using clinical images. The mean correction time of the QA system was measured. Result: ResNet-50 demonstrated higher performance in the classification. The original CNN was preferred in the regression. The orientation, angle, and left–right reversal of all images were fully corrected in all images. Moreover, patients’ positions were estimated with 96% accuracy. The mean correction time was approximately 0.4 s. Conclusion: The DL-based QA system quickly and accurately corrected CXR images.

Objective: Although cervical spondylotic myelopathy (CSM) can be easily diagnosed using magnetic resonance imaging (MRI), prediction of surgical effect using preoperative radiological examinations remains difficult. In previous studies, it was reported that diffusion tensor imaging (DTI) may be used for the prediction of surgical effect; however, these studies did not consider the influences of spinal cord compression even though the values of DTI indexes can be distorted by compressive lesions in patients with CSM. Therefore, it is uncertain whether preoperative DTI indexes can actually predict the surgical effect. The aim of this study was to investigate DTI metrics that are hardly affected by spinal cord compression and can accurately predict neurological status after decompressive surgery.Methods: Twenty-one patients with CSM who underwent surgery and 10 healthy volunteers were enrolled in this study. The subjects underwent cervical MRI, and values of DTI indexes including axial diffusivity (AD), radial diffusivity (RD), apparent diffusion coefficient (ADC), and fractional anisotropy (FA) were recorded at each intervertebral level. Further, the Japanese Orthopaedic Association (JOA) score of each patient with CSM was recorded before and after surgery for neurological status evaluation. Preoperative and postoperative values of DTI indexes were compared, and correlations between preoperative DTI parameters and postoperative neurological recovery were assessed.Results: After surgery, the lesion-adjacent (LA) ratios of RD and ADC increased (p = 0.04 and p = 0.062, respectively), while the LA ratio of FA decreased (p = 0.075). In contrast, the LA ratio of AD hardly changed. A negative correlation was observed between preoperative LA ratio of AD and JOA recovery rate 6 months after surgery (r = -0.379, p = 0.091). Based on preoperative LA ratio of AD, the patients were divided into a low AD group and a high AD group, and JOA recovery rate 6 months after surgery was found to be higher in the low AD group than in the high AD group (p = 0.024).Conclusion: In patients with CSM, preoperative LA ratio of AD is seldom affected by spinal cord compression, and it negatively correlates with JOA recovery rate 6 months after surgery.

Recent studies have indicated that injuries such as muscle tears modify the microstructural integrity of muscle, leading to substantial alterations in measured diffusion parameters. Therefore, the fractional anisotropy (FA) value decreases. However, we hypothesized that soft tissue, such as muscle tissue, undergoes reversible changes under conditions of compression without fiber injury. To evaluate the FA change due to compression in muscle tissue without fiber injury. Diffusion tensor imaging (DTI) was performed on both feet of 10 healthy volunteers (mean age = 35.0 ± 10.39 years; age range = 23–52 years) using a 3.0-T magnetic resonance imaging (MRI) scanner with an eight-channel phased array knee coil. An MRI-compatible sphygmomanometer was applied to the individuals’ lower legs and individuals were placed in a compressed state. Then, rest intervals of 5 min were set in re-rest state after compression. The FA value, apparent diffusion coefficient (ADC), and eigenvalues (λ1, λ2, λ3) of the gastrocnemius and soleus muscle were measured at each state. The mean FA values increased in all muscles in a compressed state, while the mean λ3 decreased. In all muscles, significant differences were found between the rest and compressed states in terms of mean FA and λ3 (P < 0.0001). We confirmed the reversibility of the DTI metrics, which suggests that there was no muscle injury during this study. In cases of compression without fiber injury, the FA value increases, because fibers are strongly aligned in the longitudinal direction.

[Background] In computed tomography (CT) for intracranial hemorrhage (ICH), the various window settings and the continuity of slices are critical factors for accurate diagnosis. However, traditional convolutional neural networks typically accept only single-slice images. Since ICH lesions often extend across multiple slices, using only single-slice images may result in reduced diagnostic accuracy by neglecting spatial continuity. Our approach addresses this limitation by integrating multi-slice information through a 9-channel pseudo-color map. To address this limitation, we explored the use of a 9-channel pseudo-color map for the discrimination of ICH in CT. [Method] A total of 21,744 cases (normal controls: 12,862; abnormal cases: 8882) from an open dataset were utilized for model training and validation. Abnormal cases included a variety of ICHs. The 9-channel pseudo-color map was generated by combining three different window settings with three continuous slices. ResNeXt50-32x4d architecture with five-fold cross-validation used. A total of 956 clinical cases were used for model testing. [Result] A total of 558,738 images were included in the model training process. The optimal model performance metrics were as follows: accuracy: 95.92%, sensitivity: 96.37%, and specificity: 95.24%. The average processing time for each case was recorded as 3.29 s. [Conclusions] The 9-channel pseudo-color map demonstrates high accuracy in the discrimination of ICH in CT images using deep learning methodologies.