Effectiveness evaluation of different feature extraction methods for classification of covid-19 from computed tomography images: A high accuracy classification study.

Rapid diagnosis of the Covid-19 disease is the best way to prevent infection. In this paper, it is proposed to use machine learning methods to aid diagnoses quickly Covid-19 and focused on effect of several features on classification accuracy. In the proposed method 746 axial computed tomography (CT) images of the lung; 349 Covid-19 (positives) and 397 non-Covid-19 (negative) are used. Gray-level texture, shape and first order statistical features were extracted from the images. The feature vector for model training is constructed with one feature group or combination of more than one group. We then classified with Support Vector Machine, Random Forest, k-nearest neighbor and XGBoost classifier models. The hyperparameter of the models were controlled by the tuning test. Experimental results obtained with 10-fold cross-validation. The results of cross-validation verified with the additionally independent test. The best overall accuracy was 98.65% with first order statistics features classified with XGBoost. In the gray level features, the best individual results given by GLSZM as 81.25%, and the best combination result is with GLDM, GLRLM and GLSZM features as 85.52%. An important finding of this paper is that, for Covid-19 classification, the shape and first order statistics features are more valuable than gray level features. The proposed results compared with the literature studies under some Covid-19 dataset for accuracy, precision, sensitivity and F1-score metrics. Also, the literature studies which used the different Covid-19 dataset were compared with the proposed study. Our results have the significant superiority when compared with the literature studies.

Pure ZnO and composite ZnO/TiO2 catalyst plates: a comparative study for the degradation of azo dye, pesticide and antibiotic in aqueous solutions.

Photocatalytic degradations of azo dye (RR 180), pesticide (2,4-D) and antibiotic (enrofloxacin) in aqueous solutions were performed and compared by using pure ZnO and ZnO/TiO2 composite (at 1:1 ZnO to TiO2 mole ratio) catalysts in a self-supporting plate form. The plates were produced by tape casting of the constituent powder slurries and sintering at 600°C. Photocatalytic degradations of these pollutants were carried out under UVA and UVC irradiations for 120 min. Maximum degradation was obtained for 2,4-D solution using pure ZnO plates under UVC. Due to the photolysis effect, UVC wavelength yielded higher efficiency values for all the chemicals than UVA. The discrepancy in the photocatalytic performances of the pure ZnO and the ZnO/TiO2 composite plates were not found to be significant. The plates were found to be effective for the consecutive degradation tests which indicated their potentiality in extended applications.