Improved lung cancer diagnosis using modified M3D-RUN model with fuzzy active contour segmentation and LDHA mean filtering technique

This research study proposes an enhanced version of the three-dimensional recurrent-based U-Net (M3D-RUN) model for accurate lung cancer diagnosis. The proposed model incorporates a weighted adaptive mean filter method to effectively eliminate impulsive noise and a fuzzy active contour segmentation method for precise image segmentation during the pre-processing stage. Additionally, the leader of the dolphin herd algorithm (LDHA) addresses the dimensionality issue and optimizes hyper-parameter adjustment using fitness functions. Experimental findings demonstrate that the fuzzy-based M3D-RUN segmentation model outperforms other commonly used deep learning (DL) models, achieving a maximum mean dice coefficient value of 0.7228 and a median dice coefficient of 0.7556. Moreover, the model exhibits higher sensitivity, specificity, f-score, and accuracy values, indicating its potential as a valuable mechanism for lung cancer diagnosis. In conclusion, this study highlights the effectiveness of the modified M3D-RUN model, incorporating fuzzy active contour segmentation and LDHA mean filtering technique, in significantly improving lung cancer classification, thereby benefiting medical professionals and researchers in this field.