Automated Spleen Injury Detection Using 3D Active Contours and Machine Learning.

Area of Science:

• Radiology
• Medical Imaging
• Machine Learning

Background:

• Blunt abdominal trauma frequently injures the spleen.
• Computed tomography (CT) is standard for diagnosing spleen injuries.
• Manual review of CT scans for spleen injury is time-consuming.

Purpose of the Study:

• To develop and evaluate an automated machine learning method for detecting spleen injuries.
• To improve the efficiency and accuracy of spleen injury assessment in trauma patients.

Main Methods:

• Collected CT scans from trauma patients (Michigan Medicine and CIREN datasets).
• Trained and evaluated five machine learning models: random forest, naive Bayes, SVM, k-nearest neighbors ensemble, and subspace discriminant ensemble.
• Utilized 5-fold cross-validation for model training and assessed performance on a disjoint test set.

Main Results:

• Random forest model achieved the highest performance.
• Area Under the receiver operating characteristic Curve (AUC) of 0.91.
• F1 score of 0.80 on the test set.

Conclusions:

• Automated spleen injury detection using machine learning is feasible.
• This approach has the potential to expedite triage and enhance patient outcomes.
• Quantitative assessment of traumatic spleen injury can be achieved through automated methods.