Deep learning model based on CT images to predict Ki-67 expression in renal carcinoma

Area of Science:

• Radiology
• Oncology
• Artificial Intelligence

Background:

• Ki-67 expression is a key prognostic marker in renal cell carcinoma (RCC).
• Accurate preoperative assessment of Ki-67 is crucial for tailoring treatment strategies.
• Current methods for Ki-67 assessment are invasive and time-consuming.

Purpose of the Study:

• To develop and validate a preoperative prediction model for Ki-67 expression in RCC using deep learning and CT images.
• To evaluate the clinical utility of the developed model in predicting Ki-67 levels.

Main Methods:

• A retrospective analysis of 137 RCC patients' CT images and pathological data was conducted.
• The Mobilenetv3-large deep learning model was trained using CT images from different phases (plain, arterial, venous, excretion).
• Model performance was assessed using accuracy, sensitivity, specificity, F1 score, and AUC on test and clinical validation groups.

Main Results:

• The Mobilenetv3-large model utilizing venous phase CT images demonstrated the best predictive performance.
• In the clinical validation group, the model achieved an accuracy of 0.814, with sensitivity and specificity of 0.889 and 0.667, respectively.
• The model showed good robustness and a k coefficient of 0.57, indicating clinical utility.

Conclusions:

• Deep learning applied to venous phase CT images can accurately predict preoperative Ki-67 expression in RCC.
• This non-invasive approach offers a precise and expedient tool for guiding personalized therapeutic strategies in RCC management.