ENHANCED MULTI-CLASS LUNG NODULE CLASSIFICATION USING OPTIMIZED ARTIFICIAL NEURAL NETWORK

It is important to correctly classify lung nodules in order to find lung cancer early and plan treatment. It is are very hard, though, because not all nodules are the same and some of their radiological features are the same across classes. Traditional binary classifiers don’t always do a good job of capturing the complexity of different types of nodules, like benign, malignant, inflammatory, and calcified. This makes the diagnosis less accurate and increases the number of false positives and negatives, which affects how well the treatment works. This study suggests a better multi-class Artificial Neural Network (ANN) framework to help with the classification of lung nodules. The model includes the best way to get features from CT scans of the patient by using descriptors based on shapes, textures, and histograms. To improve performance and reduce the number of dimensions, the Principal Component Analysis (PCA) method is used to choose features. We use a backpropagation algorithm to teach the artificial neural network (ANN) how to work with a set of labeled lung nodules. The suggested artificial neural network (ANN) was able to correctly sort 94.7% of the 1000 CT image samples in a test dataset. It had a kappa coefficient of 0.91, a recall of 93.5%, a precision of 92.1%, and an F1-score of 92.8%. The results of the experiment showed that these results were correct. Our method always does better than other hybrid models on a number of evaluation variables.