[Effect of AI-assisted compressed sensing acceleration on MRI radiomic feature extraction and staging model performance for nasopharyngeal carcinoma].

OBJECTIVES: To evaluate the effect of artificial intelligence-assisted compressed sensing (ACS) acceleration on MRI radiomic feature extraction and performance of diagnostic staging models for nasopharyngeal carcinoma (NPC) in comparison with conventional parallel imaging (PI). METHODS: A total of 64 patients with newly diagnosed NPC underwent 3.0T MRI using axial T1-weighted (T1W), T2-weighted (T2W), and contrast-enhanced T1-weighted (CE-T1W) sequences. Both PI and ACS protocols were performed using identical imaging parameters. The total scan time for the 3 sequences in ACS group was 227 s, representing a 30% reduction from 312 s in the PI group. Eighteen first-order and 75 texture features were extracted using Pyradiomics. Intraclass correlation coefficients (ICCs) were calculated to assess the agreement between the two acceleration methods. After feature selection using the least absolute shrinkage and selection operator (LASSO), random forest regression models were constructed to distinguish early-stage (T1 and T2) from advanced-stage (T3 and T4) NPC. The diagnostic performance of the models was evaluated using the area under the receiver operating characteristic curve (AUC) and compared using the DeLong test. RESULTS: ACS-accelerated images demonstrated good radiomic reproducibility, with 86.0% (240/279) of features showing good agreement (ICC>0.75), with mean ICCs for T1W, T2W and CE-T1W sequences of 0.91±0.09, 0.89±0.13 and 0.88±0.11, respectively. The staging prediction models achieved similar AUCs for ACS and PI (0.89 vs 0.90, P=0.991). CONCLUSIONS: The MRI radiomic features extracted using ACS and PI techniques are highly consistent, and the ACS-based model shows comparable diagnostic performance to the PI-based model, but ACS significantly reduces the scan time and provides an efficient and reliable acceleration strategy for radiomics in NPC.