Impacts of Image Preprocessing Techniques on Lung Cancer Classification

The Computed Tomography (CT) scan is one of the most reliable imaging techniques radiologists employ to diagnose lung cancer. It provides a cross-sectional and more detailed visualization of the lung. Low Dose Computed Tomography (LDCT) is an effective way to reduce radiation risks associated with CT imaging. However, in practical use this will impose noise and other artifacts that account for the misdiagnosis of cancer. Image preprocessing is a necessary step before using CT scans for computer-aided lung cancer diagnosis to achieve accurate diagnostic results by reducing noise and other artifacts. Some of the key preprocessing methods for early lung cancer diagnosis are resampling, noise removal, Contrast Limited Adaptive Histogram Equalization (CLAHE), normalization, and augmentation. This work compares various image preprocessing scenarios for lung cancer classification using CT scan images. The quality of denoised images is assessed using metrics such as Peak Signal-to-Noise Ratio (PSNR), Mean Squared Error (MSE), Structural Similarity Index Measure (SSI), and Multi-scale Structural Similarity Index Measure (MSSI). Experimental results indicate that the wavelet filter outperforms other filters in terms of these metrics. ShuffleNet is a lightweight deep neural network used to evaluate the impact of various preprocessing scenarios on lung cancer classification. The results of this work demonstrate that preprocessing techniques like resampling, noise removal, CLAHE, normalization, and augmentation enhance the performance of ShuffleNet-based classification of lung cancer with an accuracy of 96.45%, precision of 100%, recall-93.83%, F1-score-96.82%, and AUC score of 99.55%.