Detection & Classification of Bone Metastasis in Bone Scintigraphy

Abstract

This thesis proposes an approach for detection and classification of bone metastasis using spatial enhancement filters considering the bone scintigraphy is presented. Bone metastasis, the spread of cancer cells to the bones, is a critical step in diagnosis of cancer, impacting patient prognosis and treatment decisions. In this study, deep neural network (DNN) models are trained using a large dataset of bone scan images. The integration of spatial enhancement filters in the preprocessing step enhances the clarity of bone scan images, enabling more precise and reliable detection. The performance of the developed models is evaluated using various evaluation parameters, including accuracy, sensitivity, specificity, and F1-score. The respective results achieved are 95.88%, 94.44%, 96.72% and 94%, similarly, dice similarity coefficient (DSC) of 0.944 in accurate detection. For catering false positive and issue of class imbalance, classification models accurately classify using computing weights for each class accurately up to accuracy of 82.54%, precision of 72.57%, recall of 74.05% and F1-score of 73.35%. This research has implications for clinical practice, providing clinicians with a valuable tool for detection and characterization of bone metastasis, ultimately leading to improved patient care and treatment outcomes.