Diagnosis model of early Pneumocystis jirovecii pneumonia based on convolutional neural network: a comparison with traditional PCR diagnostic method.

BACKGROUND: Pneumocystis jirovecii pneumonia (PJP) is an interstitial pneumonia caused by pneumocystis jirovecii (PJ). The diagnosis of PJP primarily relies on the detection of the pathogen from lower respiratory tract specimens. However, it faces challenges such as difficulty in obtaining specimens and low detection rates. In the clinical diagnosis process, it is necessary to combine clinical symptoms, serological test results, chest Computed tomography (CT) images, molecular biology techniques, and metagenomics next-generation sequencing (mNGS) for comprehensive analysis. PURPOSE: This study aims to overcome the limitations of traditional PJP diagnosis methods and develop a non-invasive, efficient, and accurate diagnostic approach for PJP. By using this method, patients can receive early diagnosis and treatment, effectively improving their prognosis. METHODS: We constructed an intelligent diagnostic model for PJP based on the different Convolutional Neural Networks. Firstly, we used the Convolutional Neural Network to extract CT image features from patients. Then, we fused the CT image features with clinical information features using a feature fusion function. Finally, the fused features were input into the classification network to obtain the patient's diagnosis result. RESULTS: In this study, for the diagnosis of PJP, the accuracy of the traditional PCR diagnostic method is 77.58%, while the mean accuracy of the optimal diagnostic model based on convolutional neural networks is 88.90%. CONCLUSION: The accuracy of the diagnostic method proposed in this paper is 11.32% higher than that of the traditional PCR diagnostic method. The method proposed in this paper is an efficient, accurate, and non-invasive early diagnosis approach for PJP.

Prototype early diagnostic model for invasive pulmonary aspergillosis based on deep learning and big data training.

BACKGROUND: Currently, the diagnosis of invasive pulmonary aspergillosis (IPA) mainly depends on the integration of clinical, radiological and microbiological data. Artificial intelligence (AI) has shown great advantages in dealing with data-rich biological and medical challenges, but the literature on IPA diagnosis is rare. OBJECTIVE: This study aimed to provide a non-invasive, objective and easy-to-use AI approach for the early diagnosis of IPA. METHODS: We generated a prototype diagnostic deep learning model (IPA-NET) comprising three interrelated computation modules for the automatic diagnosis of IPA. First, IPA-NET was subjected to transfer learning using 300,000 CT images of non-fungal pneumonia from an online database. Second, training and internal test sets, including clinical features and chest CT images of patients with IPA and non-fungal pneumonia in the early stage of the disease, were independently constructed for model training and internal verification. Third, the model was further validated using an external test set. RESULTS: IPA-NET showed a marked diagnostic performance for IPA as verified by the internal test set, with an accuracy of 96.8%, a sensitivity of 0.98, a specificity of 0.96 and an area under the curve (AUC) of 0.99. When further validated using the external test set, IPA-NET showed an accuracy of 89.7%, a sensitivity of 0.88, a specificity of 0.91 and an AUC of 0.95. CONCLUSION: This novel deep learning model provides a non-invasive, objective and reliable method for the early diagnosis of IPA.

Machine Learning for Predicting Individual Severity of Blepharospasm Using Diffusion Tensor Imaging.

Accumulating diffusion tensor imaging (DTI) evidence suggests that white matter abnormalities evaluated by local diffusion homogeneity (LDH) or fractional anisotropy (FA) occur in patients with blepharospasm (BSP), both of which are significantly correlated with disease severity. However, whether the individual severity of BSP can be identified using these DTI metrics remains unknown. We aimed to investigate whether a combination of machine learning techniques and LDH or FA can accurately identify the individual severity of BSP. Forty-one patients with BSP were assessed using the Jankovic Rating Scale and DTI. The patients were assigned to non-functionally and functionally limited groups according to their Jankovic Rating Scale scores. A machine learning scheme consisting of beam search and support vector machines was designed to identify non-functionally versus functionally limited outcomes, with the input features being LDH or FA in 68 white matter regions. The proposed machine learning scheme with LDH or FA yielded an overall accuracy of 88.67 versus 85.19% in identifying non-functionally limited versus functionally limited outcomes. The scheme also identified a sensitivity of 91.40 versus 85.87% in correctly identifying functionally limited outcomes, a specificity of 83.33 versus 83.67% in accurately identifying non-functionally limited outcomes, and an area under the curve of 93.7 versus 91.3%. These findings suggest that a combination of LDH or FA measurements and a sophisticated machine learning scheme can accurately and reliably identify the individual disease severity in patients with BSP.

Multi-energy computed tomography reconstruction using a nonlocal spectral similarity model.

Multi-energy computed tomography (MECT) is able to acquire simultaneous multi-energy measurements from one scan. In addition, it allows material differentiation and quantification effectively. However, due to the limited energy bin width, the number of photons detected in an energy-specific channel is smaller than that in traditional CT, which results in image quality degradation. To address this issue, in this work, we develop a statistical iterative reconstruction algorithm to acquire high-quality MECT images and high-accuracy material-specific images. Specifically, this algorithm fully incorporates redundant self-similarities within nonlocal regions in the MECT image at one bin and rich spectral similarities among MECT images at all bins. For simplicity, the presented algorithm is referred to as 'MECT-NSS'. Moreover, an efficient optimization algorithm is developed to solve the MECT-NSS objective function. Then, a comprehensive evaluation of parameter selection for the MECT-NSS algorithm is conducted. In the experiment, the datasets include images from three phantoms and one patient to validate and evaluate the MECT-NSS reconstruction performance. The qualitative and quantitative results demonstrate that the presented MECT-NSS can successfully yield better MECT image quality and more accurate material estimation than the competing algorithms.

[Adaptive regularized super-resolution reconstruction for magnetic resonance images].

OBJECTIVE: To increase the resolution and signal-to-noise ratio (SNR) of magnetic resonance (MR) images, an adaptively regularized super-resolution reconstruction algorithm was proposed and applied to acquire high resolution MR images from 4 subpixel-shifted low resolution images on the same anatomical slice. The new regularization parameter, which allowed the cost function of the new algorithm to be locally convex within the definition region, was introduced by the piori information to enhance detail restoration of the image with a high frequency. The experiment results proved that the proposed algorithm was superior to other counterparts in achieving the reconstruction of low-resolution MR images.

[Adaptive super resolution algorithm for under-sampled images].

OBJECTIVE: A new algorithm of adaptive super-resolution (SR) reconstruction based on the regularization parameter is proposed to reconstruct a high-resolution (HR) image from the low-resolution (LR) image sequence, which takes into full account the inaccurate estimates of motion error, point spread function (PSF) and the additive Gaussian noise in the LR image sequence. We established a novel nonlinear adaptive regularization function and analyzed experimentally its convexity to obtain the adaptive step size. This novel algorithm can effectively improve the spatial resolution of the image and the rate of convergence, which is verified by the experiment on optical images.