Deep Neural Network for Diagnosis of Bone Metastasis

• Main Authors: Magboo, Vincent Peter C, Abu, Patricia Angela R
• Format: text
• Published: Archīum Ateneo 2022
• Subjects: Computer aided diagnosis; Convolution; Convolutional neural networks; Deep neural networks; Diseases; Medical imaging; Network architecture; Pathology; Patient treatment; Bone destruction; Bone metastasis; Bone scintigram; CNN models; Convolutional neural network; Early diagnosis; Images classification; Median survival; Quality of life; Treatment plans; Image classification; Biomedical; Computer Engineering; Electrical and Computer Engineering; Engineering; Medicine and Health Sciences; Neoplasms; Oncology
• Online Access: https://archium.ateneo.edu/discs-faculty-pubs/337; https://doi.org/10.1145/3520084.3520107
• Institution: Ateneo De Manila University

The presence of bone metastasis represents an advanced stage of malignancy with a median survival of a few months and with limited appropriate therapies. The consequent structural bone destruction leads to considerable morbidity, including untreatable pain, fractures, functional impairment which impact on the patient's quality of life. Hence, it is important to make an early diagnosis of bone metastasis to provide an accurate patient's treatment plan to improve overall survival rates and/or quality of life. The aim of this study is to develop a deep learning model using a convolutional neural network to assess the presence of bone metastasis from bone scintigram dataset of a local medical institution. The creation of the network architecture was made using an exploratory process combined with bibliographic search. Several experiments were made to determine optimum combination of parameters (input pixel size, dropout rates, batch size, and number of dense nodes). The model was also compared to the pre-trained architecture used in medical image classification reported in the literature: (1) VGG16, (2) ResNet50, (3) DenseNet121, and (4) InceptionV3. Results showed our base CNN model with good metric performance of 83.97% accuracy, 75.55% precision, 70.83% recall, 73.11% F1 score, and 89.81 % specificity. Our base CNN model outperformed VGG16, InceptionV3 and ResNet50. DenseNet121 showed the higher accuracy and precision results for this dataset, but our base CNN obtained better recall score. Our study showed promising results which could be integrated in the clinical routine workflow. The study has the potential to enhance cancer metastasis detection and monitoring.