Research on Medical Imaging Diagnosis Based on DenseNet and ResNet Model

Medical imaging diagnosis is a crucial task in the field of medicine. Among a variety of deep learning methods, convolutional neural networks have achieved remarkable progress in this domain. This investigation commences with the utilization of two distinct datasets: the COVID-19 dataset and the breast cancer dataset. Augmentation of image data is accomplished through the stochastic deletion of pixels, with subsequent replacement by mean values, resulting in the generation of augmented data samples. This procedure, which entails the input of meticulously annotated datasets to engender novel data samples, is designed to procure extensive, high-caliber annotated data. Thereafter, the ResNet-50 convolutional neural network model and the DenseNet201 dense neural network model are employed for computational analysis. In order to augment model precision and mitigate the risk of overfitting, the Dropout technique and the incorporation of a singular convolutional layer architecture are applied to the aforementioned models. These methodological refinements have been found to markedly enhance the models' fitting capabilities and overall robustness. In conclusion, the optimized models exhibited superior performance on both datasets with respect to four distinct metrics. The research approach employed herein has garnered positive outcomes, thereby underscoring its potential for continued exploration and advancement within the field.