Analysis of Optimal Model with Convolutional Neural Network and Differential Evolutionary Algorithm for Lungs Cancer Detection

Abstract

Optimizers play a pivotal role in constructing an efficient classification model. This article employs a popular deep-learning model paired with a metaheuristic—specifically, the differential evolutionary algorithm—for the crucial task of detecting lung nodules, a lethal aspect of lung cancer, a life-threatening disease. Timely treatment significantly contributes to increased survival rates, necessitating proper care and early diagnosis. To address these challenges, the Differential Evolutional Convolutional Neural Network (DECNN) emerges as the optimal solution. While Convolutional Neural Networks (CNNs) consistently yield superior results in medical applications, the intricate task of hyperparameter tuning poses a considerable challenge. Traditional optimizers such as genetic algorithms, particle swarm optimization, and random search optimization have been utilized by researchers. Differential Evolution (DE), characterized by a minimal set of parameters including population size, crossover, and mutation factors, stands out as a simple yet effective optimizer. The proposed model was implemented and tested on the IQ-OTH/NCCD datasets. To comprehensively evaluate the performance of the DECNN model optimized by the differential evolutionary optimizer, an initial model was generated and tested without the application of any optimization techniques. Subsequently, the performance and optimization criteria of this baseline model were also assessed using Genetic Algorithm and Particle Swarm Optimizer for a thorough comparative analysis.