Deep Learning for Pneumonia Detection: Leveraging SqueezeNet and Feature Extraction algorithm

Clinical experts are actively searching for accurate and efficient diagnostic methods to combat the rising cases of pneumonia. Utilizing medical imaging, such as X-ray chest radiography, for pneumonia diagnosis can be a potential solution. The main aim of the research is to investigate the potential of utilizing the Transfer learning framework for preliminary diagnosis based on X-ray radio graphs. The research incorporates a data set consisting of 10292 normal and 3896 pneumonia X-ray radio graphs to optimize the Transfer Learning framework and ensure accurate diagnosis. Data augmentation techniques are employed to enhance the data set prior to training the Transfer Learning framework based on the SqueezeNet architecture. The SqueezeNet model is trained by extracting informative features from the images during the phase of Feature extraction. The features of X-ray radio graphs are then transformed into a feature matrix, containing the x-ray parameters represented by the edge count of edge detection. To evaluate the proposed model’s performance, the feature matrix is subjected to testing using K-nearest neighbour classifiers, assessing various quality metrics including recall, F1 score, precision, area under the curve, and accuracy. The results indicate promising performance, with the Cosine KNN classifier demonstrating the best outcomes. Consequently, the proposed methodology shows potential as an initial diagnostic tool for pneumonia.