DSSAE: Deep Stacked Sparse Autoencoder Analytical Model for COVID-19 Diagnosis by Fractional Fourier Entropy

( Aim ) COVID-19 has caused more than 2.28 million deaths till 4/Feb/2021 while it is still spreading across the world. This study proposed a novel artificial intelligence model to diagnose COVID-19 based on chest CT images. ( Methods ) First, the two-dimensional fractional Fourier entropy was used to extract features. Second, a custom deep stacked sparse autoencoder (DSSAE) model was created to serve as the classifier. Third, an improved multiple-way data augmentation was proposed to resist overfitting. ( Results ) Our DSSAE model obtains a micro-averaged F1 score of 92.32% in handling a four-class problem (COVID-19, community-acquired pneumonia, secondary pulmonary tuberculosis, and healthy control). ( Conclusion ) Our method outperforms 10 state-of-the-art approaches.

Research on a Convolution Kernel Initialization Method for Speeding Up the Convergence of CNN

This paper presents a convolution kernel initialization method based on the local binary patterns (LBP) algorithm and sparse autoencoder. This method can be applied to the initialization of the convolution kernel in the convolutional neural network (CNN). The main function of the convolution kernel is to extract the local pattern of the image by template matching as the target feature of subsequent image recognition. In general, the Xavier initialization method and the He initialization method are used to initialize the convolution kernel. In this paper, firstly, some typical sample images were selected from the training set, and the LBP algorithm was applied to extract the texture information of the typical sample images. Then, the texture information was divided into several small blocks, and these blocks were input into the sparse autoencoder (SAE) for pre-training. After finishing the training, the weight values of the sparse autoencoder that met the statistical features of the data set were used as the initial value of the convolution kernel in the CNN. The experimental result indicates that the method proposed in this paper can speed up the convergence of the network in the network training process and improve the recognition rate of the network to an extent.

Concentrated Thermomics for Early Diagnosis of Breast Cancer

Thermography has been employed broadly as a corresponding diagnostic instrument in breast cancer diagnosis. Among thermographic techniques, deep neural networks show an unequivocal potential to detect heterogeneous thermal patterns related to vasodilation in breast cancer cases. Such methods are used to extract high-dimensional thermal features, known as deep thermomics. In this study, we applied convex non-negative matrix factorization (convex NMF) to extract three predominant bases of thermal sequences. Then, the data were fed into a sparse autoencoder model, known as SPAER, to extract low-dimensional deep thermomics, which were then used to assist the clinical breast exam (CBE) in breast cancer screening. The application of convex NMF-SPAER, combining clinical and demographic covariates, yielded a result of 79.3% (73.5%, 86.9%); the highest result belonged to NMF-SPAER at 84.9% (79.3%, 88.7%). The proposed approach preserved thermal heterogeneity and led to early detection of breast cancer. It can be used as a noninvasive tool aiding CBE.