Features Engineering-Driven Deep Learning Approach for Improved Pulmonary Nodules Diagnosis

Lung cancer is the most frequent cancer kind globally and the main cause of death, highlighting the crucial need for rapid and accurate diagnosis methods. Computer-aided diagnostic systems have emerged as a critical innovation in healthcare, significantly increasing both the speed and accuracy of diagnoses, However, utilizing extensive data volumes in these systems and complex deep learning models demands substantial resources and consumes significant time. This study aims to offer a novel model for diagnosing lung cancer tumors using a hybrid learning approach to reduce complexity and improve training efficiency. Within this framework, Densenet201 is employed to extract features, followed by Principal Component Analysis (PCA) for reducing dimensionality, and ultimately, feature selection is performed. The classification is then performed using a Multilayer Perceptron. The assessment utilizing the LIDC-IDRI dataset shows commendable performance during both the training and validation phases, obtaining an accuracy rate of 97.93% in training and 83.41% in validation, with an execution duration of 26 s. Future efforts will center on refining the model by increasing its effectiveness and keeping its complexity low.