Related Experiment Videos
How well can radiologists using neural network software diagnose pulmonary embolism?
J A Scott1, E L Palmer, A J Fischman
1Department of Radiology, Division of Nuclear Medicine, Massachusetts General Hospital and Harvard Medical School, Boston, MA 02114, USA.
AJR. American Journal of Roentgenology
|August 1, 2000
Summary
Artificial neural networks (ANNs) show comparable performance to experienced clinicians in diagnosing pulmonary embolism using ventilation-perfusion scans. Combining human and computer interpretations offers the best diagnostic prediction for pulmonary embolism.
Area of Science:
- Radiology
- Artificial Intelligence
- Medical Imaging
Background:
- Pulmonary embolism (PE) diagnosis relies on expert interpretation of ventilation-perfusion (V/Q) lung scans.
- Automated interpretation tools can potentially improve diagnostic accuracy and efficiency.
Purpose of the Study:
- To evaluate and optimize an artificial neural network (ANN) for diagnosing PE on V/Q scans.
- To compare ANN performance against experienced human interpreters.
Main Methods:
- 100 patients with normal chest radiographs underwent V/Q scanning and pulmonary angiography.
- Multiple ANNs were trained and their interpretations compared to three experienced nuclear medicine practitioners.
Main Results:
- ANNs performed similarly to human experts in detecting PE.
- Both human and AI interpreters performed best on large emboli.
- Optimal ANN performance was achieved with specific training strategies (e.g., right/left lung independent training).
- Collaborative interpretation (human + AI) yielded the best diagnostic predictions.
Conclusions:
- ANNs can interpret V/Q scans for PE diagnosis comparably to experienced observers.
- Human interpretation accuracy can be enhanced by integrating AI-generated predictions.
- Network training methodology is crucial for optimizing ANN performance in PE diagnosis.