A Hybrid VDV Model for Automatic Diagnosis of Pneumothorax Using Class-Imbalanced Chest X-Rays Dataset

Pneumothorax, a life-threatening disease, needs to be diagnosed immediately and efficiently. The prognosis, in this case, is not only time-consuming but also prone human errors. So, an automatic way of accurate diagnosis using chest X-rays the utmost requirement. To date, most available medical image datasets have class-imbalance (CI) issue. main theme study solve problem along with proposing automated detecting pneumothorax. find optimal approach for CI problem, we first compare existing approaches that under-bagging method (referred as data-level-ensemble formed by creating subsets majority class then combining each subset all samples minority class) outperforms other approaches. After selection best propose novel framework, named VDV model, pneumothorax detection from highly imbalance dataset. proposed model complex model-level ensemble data-level-ensembles uses three convolutional neural networks (CNN) including VGG16, VGG-19, DenseNet-121 fixed feature extractors. In data-level-ensemble, features extracted one pre-defined CNN architectures are fed support vector machine (SVM) classifier, output calculated voting method. Once outputs (corresponding different extractor) obtained, then, again, used calculate final prediction. Our framework tested on SIIM ACR Pneumothorax dataset Random Sample NIH Chest X-ray (RS-NIH). For dataset, 85.17% Recall 86.0% Area under Receiver Operating Characteristic curve (AUC) attained. second 90.9% 95.0% AUC achieved random split data while 85.45% recall 77.06% obtained patient-wise data. comparison our results both related work proves effectiveness detection.