Automatic identification of suspicious bone metastatic lesions in bone scintigraphy using convolutional neural...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Bone metastases are a common complication of cancer.
• Accurate identification of bone lesions is crucial for patient management.
• Whole-body bone scintigraphy (WBS) is a standard imaging technique for detecting bone metastases.

Purpose of the Study:

• To develop and evaluate an artificial intelligence (AI) model for identifying suspicious bone metastatic lesions.
• To utilize convolutional neural networks (CNNs) for analyzing whole-body bone scintigraphy (WBS) images.
• To compare the performance of the proposed CNN model against other established architectures.

Main Methods:

• Retrospective collection of 14,972 bone lesions from 3352 patients with malignancy.
• Manual delineation and annotation of lesions as benign or malignant by physicians.
• Evaluation of the CNN model's differentiating performance using fivefold cross-validation.
• Subgroup analyses based on lesion burden and tumor type.

Main Results:

• The proposed CNN model achieved the highest average accuracy (81.23%) in identifying suspicious bone lesions.
• The model demonstrated an average sensitivity of 81.30% and specificity of 81.14%.
• Area under the receiver operating characteristic curve (AUC) values were high across different lesion burdens and for major cancer types (lung, prostate, breast).

Conclusions:

• The developed CNN model shows significant potential for accurately identifying suspicious benign and malignant bone lesions.
• AI-guided analysis of WBS images can enhance the detection of bone metastases.
• This technology could improve diagnostic accuracy and patient care in oncology.