Analysis of Batch Size in the Assessment of Bone Metastasis from Bone Scans in Various Convolutional Neural Networks

• Main Authors: Magboo, Vincent Peter C., Abu, Patricia Angela R
• Format: text
• Published: Archīum Ateneo 2023
• Subjects: Batch size; Bone metastasis; Convolutional neural network; Biomedical Engineering and Bioengineering; Computer Engineering; Electrical and Computer Engineering; Engineering
• Online Access: https://archium.ateneo.edu/discs-faculty-pubs/363; https://doi.org/10.1007/978-981-99-3068-5_20
• Institution: Ateneo De Manila University

A diagnosis of bone metastasis indicates an advanced cancer stage with a median survival of a few months and presenting with limited therapeutic options. Hence, it is crucial to make an early assessment of bone metastasis to determine the appropriate therapeutic measure to diminish the risks of skeletal adverse events impacting on the survival rates and quality of life of patients. The objectives of this study are to develop a base convolutional neural network classifier and to determine the batch size that generate the best prediction performance for bone metastasis using bone scans. Several experiments to determine optimum batch size were made for both the base and pre-trained models. Results showed that all models favored the use of smaller batch size as larger batch sizes did not yield better performance. The findings showed the best base and pre-trained models obtained very good accuracy, precision, and superior specificity. ResNet50 bested all other models as to performance using Matthew’s correlation coefficient. However, all models had similar metric values suggesting that any of them can be used as a decision support tool for doctors in their clinical practice. Coupled with the known excellent sensitivity of bone scan as an imaging modality for bone metastasis, use of these models with superior specificity, very good precision and accuracy indicates their clinical utility leading to an enhanced diagnostic accuracy of bone scans.