Artificial intelligence-based analysis of whole-body bone scintigraphy: The quest for the optimal deep learning...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Whole-body bone scintigraphy (WBS) is crucial for early diagnosis of malignant bone diseases.
• WBS interpretation is time-consuming, subjective, and prone to errors, especially in early disease stages.
• Automating WBS analysis can improve diagnostic efficiency and accuracy.

Purpose of the Study:

• To develop and evaluate deep learning (DL) models for automated WBS interpretation.
• To classify WBS scans as normal or abnormal.
• To differentiate between malignant and non-neoplastic bone diseases using DL models.

Main Methods:

• Utilized data from 3772 and 2248 patients for two distinct analyses.
• Trained and tested ten Convolutional Neural Network (CNN) models on single- and dual-view WBS images.
• Employed feature aggregation methods like squeeze-and-excitation (SE), spatial pyramid pooling (SPP), and attention-augmented (AA) for dual-view models.
• Compared AI model performance against three nuclear medicine physicians (NMPs) using ROC curve analysis.

Main Results:

• DenseNet121_AA and InceptionResNetV2_SPP models achieved the highest performance (AUC=0.72) for classification and disease discrimination, respectively.
• AI models significantly outperformed human observers in classifying scans as normal/abnormal.
• AI performance was comparable to human observers in differentiating malignant from non-neoplastic diseases, but with drastically reduced assessment time.

Conclusions:

• Developed DL models represent a significant advancement in optimizing WBS interpretation.
• AI has the potential to assist physicians in WBS image assessment.
• Further training with larger, diverse datasets can enhance AI capabilities in nuclear medicine.