Date of Award

May 2021

Degree Type


Degree Name

Doctor of Philosophy



First Advisor

Zeyun Yu

Committee Members

Zeyun Yu, Roshan M Dsouza, Jun Zhang, Lingfeng Wang, Yi Hu


Convolutional Neural Networks, Deep learning, Ensemble classifier, Transfer learning, Wound image classification


Artificial Intelligence (AI) includes subfields like Machine Learning (ML) and DeepLearning (DL) and discusses intelligent systems that mimic human behaviors. ML has been used in a wide range of fields. Particularly in the healthcare domain, medical images often need to be carefully processed via such operations as classification and segmentation. Unlike traditional ML methods, DL algorithms are based on deep neural networks that are trained on a large amount of labeled data to extract features without human intervention. DL algorithms have become popular and powerful in classifying and segmenting medical images in recent years. In this thesis, we shall study the image classification problem in smartphone wound images using deep learning. Specifically, we apply deep convolutional neural networks (DCNN) on wound images to classify them into multiple types including diabetic, pressure, venous, and surgical. Also, we use DCNNs for wound tissue classification. First, an extensive review of existing DL-based methods in wound image classification is conducted and comprehensive taxonomies are provided for the reviewed studies. Then, we use a DCNN for binary and 3-class classification of burn wound images. The accuracy was considerably improved for the binary case in comparison with previous work in the literature. In addition, we propose an ensemble DCNN-based classifier for image-wise wound classification. We train and test our model on a new valuable set of wound images from different types that are kindly shared by the AZH Wound and Vascular Center in Milwaukee. The dataset has been shared for researchers in the field. Our proposed classifier outperforms the common DCNNs in classification accuracy on our own dataset. Also, it was evaluated on a public wound image dataset. The results showed that the proposed method can be used for wound image classification tasks or other similar applications. Finally, experiments are conducted on a dataset including different tissue types such as slough, granulation, callous, etc., annotated by the wound specialists from AZH Center to classify the wound pixels into different classes. The preliminary results of tissue classification experiments using DCNNs along with the future directions have been provided.