The three-dimensional weakly supervised deep learning algorithm for traumatic splenic injury detection and sequential...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Splenic injury is the most common solid visceral injury in blunt abdominal trauma.
• High-resolution abdominal CT is crucial for detection, yet injuries are sometimes overlooked.
• Deep learning (DL) shows promise in medical image analysis.

Purpose of the Study:

• To develop a 3D, weakly supervised DL algorithm for detecting splenic injury on abdominal CT.
• To utilize a sequential localization and classification approach for improved accuracy.

Main Methods:

• A dataset of 600 patients (50% with splenic injuries) from a tertiary trauma center was used.
• A two-step DL algorithm (localization and classification) was constructed and evaluated.
• Performance metrics included AUROC, accuracy, sensitivity, specificity, PPV, and NPV, with external validation.

Main Results:

• The DL model achieved an AUROC of 0.901, with accuracy, sensitivity, and specificity of 0.88, 0.81, and 0.92, respectively.
• Heatmaps correctly identified 96.3% of splenic injury sites in true positive cases.
• External validation showed a sensitivity of 0.92 and accuracy of 0.80.

Conclusions:

• The developed DL model effectively identifies splenic injury on CT scans.
• This AI tool has potential for practical application in trauma diagnosis.