No abstract

No abstract

Brain tumors located in the skull are among the health problems that cause serious consequences. Rapid and accurate detection of brain tumor types will ensure that the patient receives appropriate treatment in the early period, thus increasing the patient's chance of recovery and survival. In the literature, classification accuracies over 98% have been acquired automatically by using deep neural networks (DNN) for the brain tumor images such as glioma, meningioma, and pituitary. It is observed that researchers generally focused on achieving higher classification accuracy and therefore, they have used pre‐processing stages, augmentation processes, huge or hybrid DNN structures. These approaches have brought some disadvantages in terms of practical use of the developed methods: (i)The parameters of the pre‐processes should be carefully determined, otherwise the classification accuracy will decrease. (ii) In order to increase the classification performance, it is important to determine the coarse structure of the DNN correctly. If the DNN has many hyper‐parameters, the coarse structure will be determined in a long time. (iii) It is difficult to implement complex DNN structures or training algorithms in terms of practical use, because these methods need huge memory and high CPU computation. In this study, we have proposed a novel DNN model to increase the classification accuracy, and to decrease the number of weights in the structure, and to use less number of hyper‐parameters. We named this model, which uses a divergence‐based feature extractor, as DivFE‐v1 for short. 99.18% classification accuracy for the Figshare dataset is obtained by using the small‐sized DNN structure without any pre‐processing stage or augmentation process.

Pneumonia is one of the major diseases that causes a lot of deaths all over the world. Determining pneumonia from chest x-ray (CXR) images is an extremely difficult and important image processing problem. The discrimination of whether pneumonia is of bacterium or virus origin has also become more important during the pandemic. Automatic determination of the presence and origin of pneumonia is crucial for speeding up the treatment process and increasing the patient's survival rate. In this study, a convolutional neural network (CNN) framework is proposed for detection of pneumonia from CXR images. Two different binary CNNs and a triple CNN are used for determining: (i) normal or pneumonia, (ii) pneumonia of bacterium or virus origin, and (iii) normal or bacterial pneumonia or viral pneumonia. In this approach, CNNs are trained with Walsh functions to extract the features from CXR images, and minimum distance classifier (MDC) instead of a fully connected neural network (FCNN) is employed for classification purpose. Training with Walsh functions maintains the within-class scattering to be low, and between-class scattering to be high. Preferring the minimum distance classifier reduces the number of nodes used and also allows the network to be controlled with fewer hyper-parameters. These approaches bring some advantages to the system designed for the classification process: (i) easily determination of hyper-parameters, (ii) achieving higher classification performance, and (iii) use of fewer neurons. Proposed small-size CNN model was applied to CXR images from one to five years old children provided by the Guangzhou Women's and Children's Medical Center (GWCMC). Three experiments have been conducted to improve the classification performance: (i) the effect of different sizes of input images on the performance of the network was investigated, (ii) training set was augmented by randomly flipping left to right or down to up, by adding Gaussian noise to the images, by creating negative images randomly, and by changing image brightness randomly (iii) instead of RGB CXR images, gray component of the original image and four 2D wavelet images were given as input to the network. In these experiments, no major changes were observed in the classification results obtained by using the proposed CNNs. Proposed method has achieved 100% accuracy for normal or pneumonia, 92% for pneumonia of bacterium or virus origin, and 90% for normal or bacterial pneumonia or viral pneumonia. It is observed that higher classification performances were obtained with approximately five times less parameters compared to the networks that gave the best results in the literature. Thus, the applied CNN model is promising in medicine and can help experts make quick and accurate decisions.

Covid-19 infection, which first appeared in Wuhan, China in December 2019, affected the whole world in a short time like three months. The disease caused by the virus called SARS-CoV-2 affects many organs, especially the lungs, brain, liver and kidney, and causes a large number of deaths. Early detection of Covid-19 using computer-aided methods will ensure that the patient reaches the right treatment without wasting time, and the spread of the disease will be controlled. This study proposes a solution for detecting Covid-19 using chest computed tomography (CT) scan images. Firstly, features are extracted by Xception network, convolutional neural network (CNN) based transfer learning method, then classification process is performed with a fully connected neural network (FCNN) added at the end of this architecture. The classification model was tested ten times on the accessible SARS-CoV-2-CTscan dataset containing 2482 CT images labelled as covid and non-covid. The precision, recall, f1-score and accuracy metrics were used as performance measures; and ROC curve related to the model was drawn. While obtaining an average of 98.89% accuracy, in the best case, 99.59% classification performance was achieved. Xception outperforms other methods in the literature. The results promise that the proposed method can be evaluated as a clinical option helping experts in the detection of Covid-19 from CT images.