Hybrid Deep Learning Approach with Feature Engineering for Enhanced Pulmonary Nodule Diagnosis

Lung cancer is the primary cause of mortality globally in both males and females, underscoring the urgent requirement for rapid and accurate methods of early detection for this condition. Computer-aided diagnosis systems have demonstrated their effectiveness and efficiency in the medical industry for the past twenty years. However, when working with large datasets, difficulties emerge, particularly with efficiency, especially when employing time-consuming deep learning models. This study aims to introduce a novel hybrid model that utilizes the advantages of effective deep learning models, specifically the Convolutional Neural Network, to extract a significant number of features. Subsequently, the model selects the most pertinent characteristics, reducing intricacy and enhancing effi- cacy within a shorter timeframe. This methodology utilizes the Densenet201 model to extract features and subsequently applies Principal Component Analysis (PCA) to decrease dimensionality and conduct feature selection. Afterward, a multilayer perceptron is used for categorization. Furthermore, to address the model’s explainability, Local Interpretable Model-Agnostic explanations (LIME) are used to explain the model’s individual prediction. The performance of the LIDC-IDRI dataset was exceptional during both the training and validation phases of evaluation. The accuracy rate reached 99.22