Optimized convolutional neural network for automatic lung nodule detection with a new active contour segmentation

The effectiveness of treatment for lung nodules is significantly impacted by early diagnosis of the nodules on CT scans. Radiologists are aided by the utilization of CAD technologies. The purpose of this paper is to present a methodology for identifying lung cancer: "(i) pre-processing, (ii) segmentation, (iii) feature extraction, and (iv) detection." The GF methods commence with the input image being pre-processed. To segment the pre-processed images, a "improved cross-entropy-based active contour segmentation model" is implemented. Then, using a CNN that has been optimized, features such as LBP, entropy, and contrast are computed. The self-adaptive tunicate swarm algorithm (SATSA) is used to optimize CNN weights. For this study, the LIDC-IDRI dataset was used, and Python was used for experimentation. Because pulmonary nodules can differ significantly in size, shape, texture, and appearance, it's crucial to include a wide variety of patterns in the training dataset. The authors hope to improve the generalization and robustness of the CNN model in identifying lung nodules by incorporating a large number of patterns. Comparing the CNN + SATSA model to existing models allowed researchers to assess its effectiveness in identifying lung nodules. The suggested model's correctness is demonstrated by its high recall, false discovery rate, FNR, MCC, FPR, precision, FOR, accuracy, specificity, NPV, FMS, and sensitivity values.