Deep learning enabled fast free-breathing stack-of-stars multiparameter mapping for fully quantitative analysis of prostate carcinoma.

Motivation: The diagnosis of prostate cancer continues to be highly qualitative, resulting in reduced diagnostic precision. Goal(s): We present fast, quantitative transient-state T1-T2 mapping of the prostate with enhanced image quality on a larger prostate cancer patient cohort, along with an evaluation of quantitative parameters on tissue and lesion heterogeneities of the prostate. Approach: (i) Incorporation of deep learning reconstruction with quantitative transient-state imaging combined with stack-of-stars based acquisition. (ii) Quantitative parameter analysis using unsupervised K-means clustering. Results: (i) Significant improvement of SNR and image quality using deep learning reconstruction. (ii) Identification of different tissues and lesions using quantitative T1-T2 relaxation times. Impact: Quantitative transient-state imaging combined with deep learning-based reconstruction provides high image quality T1 and T2 maps, enabling a fully quantitative evaluation of prostate cancer diagnosis with the potential of improving the prostate diagnosis pipeline.