Interactive segmentation for accurately isolating metastatic lesions from low-resolution, large-size bone...

Area of Science:

• Medical Imaging
• Oncology
• Computer-Aided Diagnosis

Background:

• Bone metastases are common in malignant tumors, necessitating accurate detection and delineation in SPECT imaging for treatment planning.
• Current manual delineation of bone lesions in SPECT scans is subjective and prone to variability.
• Automated segmentation methods often yield high false positive rates, limiting clinical application.

Purpose of the Study:

• To develop an interactive segmentation framework for improving the accuracy of bone metastasis delineation in SPECT images.
• To address the challenges of segmenting low-resolution, large-size SPECT bone scans.
• To create a prototype tool for clinical assistance in bone metastasis segmentation.

Main Methods:

• An interactive segmentation framework utilizing deep convolutional neural networks was proposed.
• A U-shaped backbone network was incorporated to handle inter-patient variability.
• An interactive attention module was employed to enhance feature extraction in dense bone regions.

Main Results:

• The proposed framework demonstrated effectiveness through extensive experiments on clinical data.
• A prototype tool was successfully developed to assist in segmenting metastatic bone lesions.
• The framework supports the creation of a large-scale dataset for bone metastasis segmentation.

Conclusions:

• The interactive segmentation framework effectively segments bone metastases in SPECT scans, particularly for lung cancer.
• The developed prototype tool shows clinical application value in assisting with bone metastasis segmentation.
• This approach offers a promising solution for challenging SPECT bone scan segmentation tasks.