Prediction of muscle invasion in upper urinary tract urothelial carcinoma using 3D reduced-FOV CUBE T2 with deep learning reconstruction

Motivation: Accurate imaging assessment of muscle invasion in upper urinary tract urothelial carcinoma (UTUC) is crucial for surgical planning. Goal(s): To better display the anatomy between tumor and adjacent urinary wall using 3D reduced-FOV (rFOV) CUBE T2 with deep learning reconstruction (DLR), thus improving muscle invasion prediction in UTUC. Approach: The structural clarity of CUBE T2 images with and without DLR were scored and compared. The irregular surface of tumor attachment was assessed using a novel evaluation method. Results: CUBE T2 with DLR better depicted irregular surfaces of tumor attachment. The proposed method showed promising specificity and negative predictive value for muscle invasion prediction. Impact: The visualization of irregular attachment surfaces using 3D rFOV CUBE T2 with DLR offers a new perspective for predicting muscle invasion in UTUC. This could assist patients without irregular attachment surfaces in avoiding radical nephroureterectomy.