Deep learning system for malignancy risk prediction in cystic renal lesions: a multicenter study

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Renal Pathology

Background:

• Cystic renal lesions (CRLs) are increasingly detected, necessitating accurate malignancy risk stratification.
• Distinguishing benign from malignant CRLs is crucial for appropriate patient management.
• Current diagnostic methods can be invasive or lack definitive accuracy.

Purpose of the Study:

• To develop and validate an interactive, non-invasive artificial intelligence (AI) system for predicting malignancy risk in CRLs.
• To enhance diagnostic accuracy and clinical decision-making for CRLs.
• To reduce overtreatment and excessive follow-up associated with incidental CRLs.

Main Methods:

• A retrospective, multicenter study involving 715 patients with CRLs.
• Development of a 3D segmentation model for CRLs using geodesic-based methods.
• Implementation of a spatial encoder temporal decoder (SETD) classification model combining 3D-ResNet50 and gated recurrent unit (GRU) for multi-phase CT analysis.
• Performance evaluation using metrics such as AUC, accuracy, Dice similarity, IOU, sensitivity, and specificity.

Main Results:

• The AI system achieved excellent performance in both validation and testing datasets.
• Validation dataset: AUC=0.973, Accuracy=0.916, Dice=0.847, IOU=0.743, Sensitivity=0.840, Specificity=1.000.
• Testing dataset: AUC=0.998, Accuracy=0.988, Dice=0.861, IOU=0.762, Sensitivity=0.876, Specificity=1.000.

Conclusions:

• The AI system demonstrates strong ability to differentiate between benign and malignant CRLs.
• The developed system shows significant potential for improving clinical decision-making in CRL management.
• This non-invasive AI tool offers a promising solution for accurate CRL diagnosis in an era of prevalent incidental findings.