Abstract
BACKGROUND: Multiparametric MRI has shown great promise to derive multiple quantitative imaging biomarkers for treatment response assessment. PURPOSE: To evaluate a novel deep-learning-enhanced MUlti-PArametric MR sequence (DL-MUPA) for treatment response assessment for brain metastases patients treated with stereotactic radiosurgery (SRS) and head-and-neck (HN) cancer patients undergoing conventionally fractionation adaptive radiation therapy. METHODS: DL-MUPA derives quantitative T1 and T2 relaxation time maps from a single 4-6-min scan denoised via DL method using least-squares dictionary fitting. Longitudinal phantom benchmarking was performed on a NIST-ISMRM phantom over 1 year. In patients, longitudinal DL-MUPA data were acquired on a 1.5T MR-simulator, including pretreatment (PreTx) and every ∼3 months after SRS (PostTx) in brain, and PreTx, mid-treatment and 3 months PostTx in HN. Delta analysis was performed calculating changes of mean T1 and T2 values within gross tumor volumes (GTVs), residual disease (RD, HN), parotids, and submandibular glands (HN) for treatment response assessment. Uninvolved normal tissues (normal appearing white matter in brain, masseter in HN) were evaluated for within-subject repeatability. RESULTS: Phantom benchmarking revealed excellent inter-session repeatability (coefficient of variation < 0.9% for T1, < 6.6% for T2), suggesting reliability for longitudinal studies with systematic bias adjustment. Uninvolved normal tissue suggested acceptable within-subject repeatability in the brain |ΔT1(mean)| < 36 ms (4.9%), |ΔT2(mean)| < 2 ms (6.1%) and HN |ΔT1(mean)| < 69 ms (7.0%), |ΔT2(mean)| < 4 ms (17.8%) with few outliers. In brain, remarkable changes were noted in a resolved metastasis (4-month PostTx ΔT1(mean )= 155 ms (13.7%)) and necrotic settings (ΔT1(mean )= 214-502 ms (17.6-39.7%), ΔT2(mean )= 7-41 ms (8.7-41.4%), 6-month to 3-month PostTx). In HN, two base of tongue tumors exhibited T2 enhancement (PostTx GTV ΔT2(mean )> 7 ms (12.8%), RD ΔT2(mean )> 10 ms (18.1%)). A case with nodal disease resolved PostTx (GTV ΔT1(mean )= -541 ms (-39.5%), ΔT2(mean )= -24 ms (-32.7%), RD ΔT1(mean )= -400 ms (-29.2%), ΔT2(mean )= -25 ms (-35.3%)). Parotids (PostTx ΔT1(mean )> 82 ms (12.4%), ΔT2(mean )> 6 ms (13.4%)) and submandibular glands (PostTx ΔT1(mean )> 135 ms (14.6%), ΔT2(mean )> 17 ms (34.5%)) adjacent to gross disease exhibited enhancement while distant organs remained stable. CONCLUSIONS: Preliminary results suggest promise of DL-MUPA for treatment response assessment and highlight potential endpoints for functional sparing.