Deep Learning Analysis With Gray Scale and Doppler Ultrasonography Images to Differentiate Graves’ Disease
Authors' conclusion
Machine learning models based on ultrasound and Doppler images showed promising results with high accuracy and sensitivity in differentiating GD from thyroiditis.
Study context
- Population
- Patients initially presenting with thyrotoxicosis who underwent thyroid US at a single tertiary hospital: 351 with Graves' disease (GD) (7719 images) and 136 with thyroiditis (2980 images)
- Comparison
- Xception vs EfficientNetB0_2 deep learning models for differentiating GD from thyroiditis
- Outcome
- Diagnostic performance for GD (accuracy, sensitivity, specificity, PPV, NPV, F1 score)
Methodological detailssingle tertiary hospital, deep learning, data augmentation, Image preprocessing, model generation and training verification on DEEP:PHI® platform
- Single tertiary hospital
- Deep learning
- Data augmentation
- Image preprocessing, model generation and training verification on DEEP:PHI® platform
Performance
As reported in the abstract
Diagnosis of GD
| Test or model | Sensitivity | Specificity | PPV | NPV | Accuracy | F1 score |
|---|---|---|---|---|---|---|
| Xception | 89.26% | 73.17% | 90.06% | 71.43% | 84.94% | 89.66 |
| EfficientNetB0_2 | 90.28% | 71.78% | 89.71% | 73.05% | 85.31% | 89.99 |
Sensitivity
Xception89.26%
EfficientNetB0_290.28%
Specificity
Xception73.17%
EfficientNetB0_271.78%
PPV
Xception90.06%
EfficientNetB0_289.71%
NPV
Xception71.43%
EfficientNetB0_273.05%
Accuracy
Xception84.94%
EfficientNetB0_285.31%
F1 score
Xception89.66
EfficientNetB0_289.99
Original publication
Deep Learning Analysis With Gray Scale and Doppler Ultrasonography Images to Differentiate Graves’ Disease. J Clin Endocrinol Metab. Published 13 April 2024. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →