Computer-aided detection of bone metastasis in bone scintigraphy images using parallelepiped classification method.

Area of Science:

• Nuclear Medicine
• Medical Imaging Analysis
• Machine Learning in Healthcare

Background:

• Accurate diagnosis of bone metastases is crucial for treatment planning.
• Existing automated systems for bone metastasis detection require further improvement.
• Novel interpretation methods for bone scintigraphy are needed.

Purpose of the Study:

• To present a new modality for bone metastasis detection using parallelepiped classification (PC) on bone scintigraphy images.
• To map radionuclide distribution for enhanced visualization and analysis.
• To evaluate the accuracy of the PC method in differentiating metastatic bone from normal tissue.

Main Methods:

• Bone scintigraphy images from 12 patients with bone metastases were analyzed.
• Parallelepiped classification (PC) was used to generate color maps of radionuclide distribution.
• Seven classes of radionuclide accumulation were identified and processed using machine learning software.
• Accuracy was assessed using statistical measurements, including a confusion matrix, overall accuracy, producer's and user's accuracies, and the κ coefficient.

Main Results:

• The PC method demonstrated high precision in differentiating metastatic bone from normal tissue.
• Overall classification accuracy was 87.58% ± 2.25%, with a κ coefficient of 0.8367 ± 0.0252.
• Generated color maps offered improved contrast, facilitating easier interpretation and detection of subtle differences in radionuclide accumulation.

Conclusions:

• Preliminary findings suggest that parallelepiped classification (PC) combined with bone scintigraphy is a valuable tool for bone metastasis detection.
• This approach can lead to more accurate image interpretation and improved diagnostic accuracy.
• The method has the potential to positively impact clinical decision-making regarding cancer treatment.