Early Stage Cancer Detection and Classification Using Improved Graph Convolutional Recurrent Neural Network-based Optimization
International journal of intelligent engineering and systems · 2025
Classifying cancer in its early stages is essential for prompt cancer identification as well as treatment, which has a major influence on survival rates and patient outcomes.Early cancer detection increases the likelihood related to a successful course of therapy and a full recovery by identifying the disease when it is still localized and has not spread to remaining regions of the body.Effective cancer type as well as stage classification is made possible by deep learning methods exceptional ability to recognize complex patterns as well as characteristics from a variety of sources, including clinical records, genetic data, as well as medical images.Through tailored medicines and early intervention, deep learning has the feasibility to improve patient outcomes in the early stages of cancer classification.Therefore, this research paper detects and classifies the early-stage cancer using novel intelligent deep learning methodology.Initially, the data is gathered from the standard benchmark dataset such as the kaggle.For the collected data, the pre-processing is done for removing the noise and enhancing the contrast.Here, the noise removal is accomplished by the Wiener filter and the contrast enhancement is performed by the logarithmic transformation method.These pre-processed images undergo the feature extraction and fusioning using the Neural Architecture Search Net (NASNet) and EfficientNetV2S methods.For these extracted features, the final classification takes place with the help of novel Improved Graph Convolutional Recurrent Neural Network (IGCRNN).Here, the parameter tuning of GCRNN is done using the Root Mean Square Propagation (RMSProp) optimizer with the intention of attaining the accuracy maximization as the main objective function, thus referred to be novel IGCRNN.This proposed IGCRNN classifies the final early-stage cancer output into four classes such as normal, adenocarcinoma, large cell carcinoma, and squamous cell carcinoma respectively.Simulation results show good accuracy rates in differentiating among malignant and non-cancerous cells for deep learning-based early-stage cancer classification.Robust effectiveness in analyzing various data types is demonstrated by the deep learning methods, which result in accurate classification of cancer stages, which is essential for customized treatment plans and prompt interventions.The proposed IGCRNN for the early-stage cancer classification model in terms of accuracy is 31.11%,14.99%, 12.78%, and 31.94%better than HDS+SDS, EL-SVM, IEWT+DL-CEWT, and HAS respectively.Similarly, the proposed IGCRNN for the early-stage cancer classification model with respect to precision is 14.86%, 33.21%, 19.97%, and 12.11% advanced than HDS+SDS, EL-SVM, IEWT+DL-CEWT, and HAS respectively.