Deep learning based classification of solid lipid-poor contrast enhancing renal masses using contrast enhanced CT.

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Accurate classification of renal masses is crucial for patient management.
• Distinguishing between benign and malignant renal lesions can be challenging.
• Multiphase contrast-enhanced CT (CECT) is a key imaging modality for renal mass evaluation.

Purpose of the Study:

• To develop and assess a workflow using convolutional neural networks (CNNs) for classifying solid, lipid-poor, contrast-enhancing renal masses.
• To evaluate the diagnostic performance of the developed CNN classifier on multiphase CECT images.

Main Methods:

• Retrospective analysis of 143 patients with predominantly solid, lipid-poor renal lesions (46 benign, 97 malignant).
• Manual segmentation of whole tumor volumes and selection of axial CT phase images.
• Training a CNN classifier using augmented and split data (83.9% training, 16.1% validation).
• Quantification of CNN performance using eightfold cross-validation.

Main Results:

• The CNN classifier achieved an overall accuracy of 78% (95% CI: 76-80%).
• Sensitivity was 70% (95% CI: 66-74%) and specificity was 81% (95% CI: 79-83%).
• The area under the curve (AUC) for the classifier was 0.82.

Conclusions:

• A CNN-based classifier was successfully developed to diagnose solid enhancing malignant renal masses.
• The study demonstrates that CNNs can be trained to accurately differentiate malignant renal lesions from benign ones using CECT imaging.