Automated Diagnosis of Bone Metastasis by Classifying Bone Scintigrams Using a Self-defined Deep Learning Model

Area of Science:

• Medical Imaging
• Artificial Intelligence in Oncology
• Nuclear Medicine

Background:

• Bone metastasis is a common complication in cancer patients, affecting bone in approximately 70% of cases.
• Early detection of bone metastases is critical for effective treatment and improved patient survival rates.
• Deep learning models are increasingly utilized for medical image analysis, showing promise in diagnostic applications.

Purpose of the Study:

• To develop and evaluate a deep learning-based automated classification model for diagnosing bone metastasis from bone scintigraphy images.
• To investigate the efficacy of a convolutional neural network for detecting lung cancer bone metastasis.

Main Methods:

• A custom convolutional neural network (CNN) was designed, comprising feature extraction and classification sub-networks.
• The model processed SPECT bone scintigrams, extracting hierarchical features for classification into metastatic or non-metastatic categories.
• Image fusion of anterior and posterior scans, excluding the urinary bladder, was employed to optimize performance.

Main Results:

• The proposed deep learning model achieved an Area Under the Curve (AUC) of 0.8489 for detecting bone metastasis.
• Optimal performance was obtained by fusing anterior and posterior scans after excluding bladder activity.
• The model demonstrated strong performance across various metrics, including accuracy (0.8038), precision (0.8051), recall (0.8039), and F-1 score (0.8036).

Conclusions:

• The developed two-class classification network effectively predicts the presence of lung cancer bone metastasis from bone scintigraphy.
• Urinary bladder activity negatively impacts automated bone metastasis diagnosis, suggesting its exclusion is beneficial.
• The deep learning model shows superior performance compared to existing classical deep learning approaches for this diagnostic task.