Deep Learning on Bone Scintigraphy to Detect Abnormal Cardiac Uptake at Risk of Cardiac Amyloidosis

Area of Science:

• Nuclear Medicine
• Artificial Intelligence in Radiology
• Cardiovascular Imaging

Background:

• Cardiac uptake on technetium-99m whole-body scintigraphy (WBS) is a key indicator for transthyretin cardiac amyloidosis.
• Misdiagnosis can occur due to unfamiliarity with this scintigraphic finding, despite characteristic images.
• Retrospective WBS analysis can identify undiagnosed patients with cardiac amyloidosis.

Purpose of the Study:

• To develop and validate a deep learning model for automatic detection of significant cardiac uptake (Perugini grade ≥2) on WBS.
• To identify patients at risk for cardiac amyloidosis using large hospital databases.
• To improve diagnostic accuracy and reduce misdiagnosis of cardiac amyloidosis.

Main Methods:

• A convolutional neural network (CNN) model was developed using image-level labels.
• Performance was evaluated using C-statistics with 5-fold cross-validation and an external validation dataset.
• The model was trained and validated on thousands of WBS images, including positive and negative cases.

Main Results:

• The model achieved high performance: 98.9% sensitivity and 99.5% specificity in cross-validation, and 96.1% sensitivity and 99.5% specificity in external validation.
• Area under the receiver-operating characteristic curve (AUC) was 0.999 in both validation schemes.
• Factors like sex, age, BMI, and acquisition parameters had minimal impact on model performance.

Conclusions:

• The developed deep learning model effectively identifies significant cardiac uptake on WBS (Perugini grade ≥2).
• This AI tool shows promise for aiding in the diagnosis of cardiac amyloidosis.
• The model can assist in identifying patients with cardiac amyloidosis from large WBS databases.