A Hybrid System: Convolutional Neural Networks with Discrete Wavelet Transform for Lung Cancer Detection

Abstract

Lung-related illnesses such as pneumonia, lung cancer, and COVID-19 are

among the major causes of death across the globe, and thus, quick and precise

diagnostic techniques are required. Traditional diagnosis using routine X-ray

and CT imaging. is time-consuming and greatly relies on the expertise of

radiologists. This paper establishes a new hybrid deep learning approach that

proposes a combination of a Discrete Legendre Wavelet Transform (DLEWT)

and a convolutional neural network (CNN) to improve the automated early lung

disease screening. The algorithm involves three steps: (1) orthogonal Legendrebased scling and wavelet functions are constructed over the interval [−1, 1]; (2)

multi-level 2D DLEWT decomposition is performed with the aim of extracting

approximation and detail coefficients corresponding to anatomical structures

and pathological features respectively, (3) threshold-based denoising and

CLAHE . The wavelet features extracted are then integrated into a CNN model

with the standard convolutional kernels substituted with DLEWT learnable

filters.