Radiomics-based machine learning for splenic injury diagnosis using computed tomography (CT) images

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Efficient trauma assessment is crucial for patient care, with imaging vital for injury detection.
• Manual CT image analysis for splenic trauma is subjective and time-consuming, necessitating automated diagnostic tools.
• Radiomics and machine learning offer objective approaches for analyzing medical images.

Purpose of the Study:

• To evaluate machine learning models and radiomics features for diagnosing traumatic spleen injuries on CT images.
• To compare the performance of different machine learning algorithms in classifying splenic trauma.
• To assess the potential of automated tools in aiding radiologists in spleen lesion detection.

Main Methods:

• A dataset of 600 CT images (mild/severe splenic injury, healthy controls) was utilized.
• Radiologist segmentation identified regions of interest for radiomics feature extraction.
• Twenty-five machine learning models were evaluated, with Light Gradient Boosting Machine, Ridge Classifier, and Adaptive Boosting selected for detailed analysis.

Main Results:

• Light Gradient Boosting Machine achieved 98% accuracy for mild spleen injuries.
• Adaptive Boosting showed 90% accuracy for severe spleen injuries.
• Selected models demonstrated high precision and specificity in diagnosing traumatic spleen lesions.

Conclusions:

• Machine learning models, particularly Light Gradient Boosting Machine and Adaptive Boosting, show significant capability in automatically detecting traumatic spleen injuries on CT scans.
• Integrating radiologist expertise with these models allows for rapid pre-screening of potential spleen lesions.
• These automated tools can enhance the efficiency and objectivity of splenic trauma assessment in clinical practice.