Venous-phase CT deep learning model predicts Ki-67 in renal cancer

A deep learning model using venous-phase CT images predicted high vs low Ki-67 expression in renal cell carcinoma with accuracy 0.814 (sensitivity 0.889, specificity 0.667) in a clinical validation set of 27 patients.

• The venous phase CT model outperformed models using other contrast phases.
• The model's agreement with pathology was moderate (κ=0.57) in clinical validation.
• The study used a separate clinical validation set of 27 patients not involved in training.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.