Clinical Report: Enhanced Lesion Characteristics and Boundary-Focused Fusion for Identifying Pulmonary Embolism in CT Imaging
Overview
This study presents an automatic recognition method for pulmonary embolism in CT images, achieving high accuracy and precision. The proposed framework combines lesion feature enhancement and boundary-related structural cue fusion.
Background
Pulmonary embolism is a critical condition characterized by the obstruction of pulmonary arteries, leading to significant morbidity and mortality. Accurate diagnosis is essential, yet the nonspecific clinical manifestations often complicate recognition. CT pulmonary angiography is a key imaging modality for diagnosis.
Data Highlights
Metric
Value
Accuracy (Internal)
0.956
Precision
0.961
Recall
0.951
AUC
0.971
Accuracy (FUMPE External)
0.779
Accuracy (RSNA External)
0.672
Key Findings
The proposed method achieved an internal accuracy of 0.956 in recognizing pulmonary embolism.
Precision and recall rates were 0.961 and 0.951, respectively.
The area under the curve (AUC) for the internal evaluation was 0.971.
In external validation, the method demonstrated accuracy values of 0.779 and 0.672 on the FUMPE and RSNA datasets, respectively.
The framework incorporates a Pulmonary Embolus Feature Enhancement Module and a Vascular Boundary Aware Fusion Module.
Clinical Implications
The findings indicate that the proposed method enhances the accuracy of pulmonary embolism detection in clinical settings.
Conclusion
The study indicates that the proposed automatic recognition method significantly improves the discriminative performance of pulmonary embolism recognition from CT images.
