Enhanced Pancreas Cancer Detection Using Dynamic ConvNets

In investigation, used a deep learning-based algorithm that is Enhanced dynamic Convolutional Neural Networks (ConvNets) for identifying the Pancreatic Ductal Adenocarcinoma (PDA) is one of the types of Pancreatic Cancer. Which is most common and aggressive form of cancer, characterized by poor diagnosis and increased mortality rates. Early detection is important for saving patient's life. Here, methodology identifies PDA as Malignant and Benign cases where Malignant refers as consists of cancerous cells whereas Benign refers as non-cancerous cells. This study uses a Convolutional Neural Network (CNN) enhanced with dynamic convolutional layers and residual blocks to offer a fresh way of classifying images from CT scans images of the pancreas. The proposed model makes use of novel DynamicConv2D layer and state-of-the-art data augmentation approaches to enable adaptive feature extraction based on input variability. A balanced class weight mechanism and a well-planned learning rate schedule are used to train the model using the AdamW optimizer in order to address the problem of class imbalance. Metrics such as ROC-AUC, F1-score, and precision-recall values are used to evaluate the effectiveness of the model and demonstrate its ability to discriminate between Malignant and Benign situations.