Radiography is an accepted technique for the medical community to detect abnormalities. However, the interpretation of the images is time- onsuming and is subject to error by radiologists who are exposed to external factors including possible fatigue from working long hours, overwork, or thinking about other life matters.
To improve the efficiency of the radiologists' work, we developed a computer-assisted diagnostic model to classify the radiographs into two classifications: a normal condition and an anomaly (or an abnormal state) in order to facilitate the radiological diagnostic process, by transferring a set of selected deep convolutional neural networks between a set of available networks. We have studied and applied them on the basis of potentially abnormal areas that radiologists provide for the study case we have selected In this study, we shed light on an experimental study of a fully connected convolutional neural network (DenseNet) by applying it to the studied data sample through a learning transfer technique.
We obtained a set of good results, which achieved high diagnostic accuracy of about 88% in some of the studied cases
Al-Baath University Journal.
2021.
