IMENet: A Deep Learning Model for Lung Cancer Detection Using Multiple Architectures

Abstract

In the modern era of artificial intelligence and medical imaging, accurate and efficient lung cancer detection remains a critical challenge. The increasing availability of large-scale medical imaging datasets has facilitated the development of AI-driven techniques for early diagnosis. The research creates IMENet as a deep learning framework by combining InceptionV3 with MobileNetV3Large and EfficientNetB7 to achieve superior features extraction and classification results. Trainings and evaluations of the model happen using data from the LUNA16 subset of LIDC/IDRI's public database. The preprocessing starts with Discrete Wavelet Transform that reduces noise then moves to Contrast Limited Adaptive Histogram Equalization and contrast stretching for intensity correction before using log transformation to enhance low intensity values. The training and testing phases contain ratios of 80:20 split after applying Synthetic Minority Over-sampling Technique (SMOTE) to correct class imbalance. The optimization of the IMENet model uses the Adam optimizer with an optimization rate of 0.00001 while implementing batches of 200 data points during 100 training epochs. Superior performance classification abilities of the model become apparent through the evaluation metrics which include accuracy, precision, recall, F1-score, sensitivity and specificity. The generated model reaches 95.5% accuracy which exceeds the performance of CNN and SVM-CNN and RF+CNN traditional models. IMENet stands out in lung cancer detection effectiveness thus showing promise for clinical medical applications. For future advancement the model will be expanded to analyze multiple categories and will implement AI interpretation methods and needs performance adjustments to achieve real-time medical diagnosis capabilities.