LUNG DISEASE DETECTION WITH K-NEAREST NEIGHBORS ALGORITHM

Abstract

Among the major causes of death and illness worldwide are respiratory illnesses such as
pneumonia, tuberculosis (TB), lung cancer, and chronic obstructive pulmonary disease
(COPD), and thus accurate and correct diagnosis is required to provide proper therapy
treatment. The traditional diagnostic procedures—X-ray imaging, computed tomography
(CT) imaging, and pulmonary function tests—depend on the experience of the clinician and
thus may lead to diagnostic discrepancy and time wastage. For automation and precision in
lung disease detection, machine learning techniques—in the form of the K-Nearest
Neighbors (KNN) algorithm—have been extensively integrated into computer-aided
diagnostic (CAD) systems in recent years. The application of the KNN algorithm for lung
disease classification using clinical factors and radiographic images is explored in this
work. X-ray and CT scan images and respective diagnostic labels constitute the model
training set. Feature extraction techniques such as the Histogram of Oriented Gradients
(HOG) and Gray Level Co-occurrence Matrix (GLCM) are utilized to enhance diseasespecific features prior to KNN classification. With a 94.2% classification rate, results demonstrate that KNN outperforms traditional rule-based diagnosis systems. Further, the
authors of the article suggest in the paper feature selection, data augmentation, and
parameter optimization to enhance model performance. Since results demonstrate, KNN
can be utilized as an effective tool for radiologists in early diagnosis as it can identify
pulmonary disorders.