A Novel Convolutional Neural Network for Medical Image Assisted Diagnosis

Early and accurate detection of dangerous diseases is critical to avoiding deadly consequences, especially in young children and the elderly. In this work, as a support for the diagnosis of diseases, we propose a novel 37-layer Convolutional Neural Network (CNN) architecture. The new designed model is trained and tested using two medical datasets (DB1 for chest X-ray and DB2 for dermoscopy melanoma skin cancer). The model is used in binary and multiple classifications, achieving robust outcomes. It was evaluated using accuracy, sensitivity, specificity, precision, recall, and F1 score. By using the CNN proposed method, the accuracy of DB1 classification was 97.1% for binary classification and 96.2% for multi-classification, the accuracy of DB2 classification was 98.2% for binary classification and 96.6% for multi-classification. The outcomes are promising, and the new CNN model can be utilized to detect lung diseases and melanoma skin cancer early with high accuracy and low cost.