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

doi:10.12122/j.issn.1673-4254.2025.11.01

Abstract

Abstract:

Objective 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). Conclusion 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.

Key words: artificial intelligence, magnetic resonance imaging, radiomics, agreement, nasopharyngeal carcinoma

Xinyang LI, Guixiao XU, Jiehong LIU, Yanqiu FENG. Effect of AI-assisted compressed sensing acceleration on MRI radiomic feature extraction and staging model performance for nasopharyngeal carcinoma[J]. Journal of Southern Medical University, 2025, 45(11): 2518-2526.

Figures/Tables 8

Tab.1 Characteristics of the patients with NPC

Characteristics	Dataset values
Number of patients (%)	64
Age (year, Mean±SD)	44.2±11.3 (18-68)
Gender
Male	42(65.6%)
Female	22(34.4%)
Histological type
NKUC	64 (100%)
T stage
T1	7 (10.9%)
T2	6 (9.4%)
T3	40 (62.5%)
T4	11 (17.2%)
N stage
N0	1 (1.6%)
N1	32 (50%)
N2	15 (23.4%)
N3	16 (25%)
M stage
M0	59 (92.2%)
M1	5 (7.8%)

Tab.2 MRI sequences and parameters

Sequences parameter	AX T2WI FSE ACS	AX T2WI FSE PI	AX T1WI FSE ACS	AX T1WI FSE PI	Post-contrast AX T1WI FSE ACS	Post-contrast AX T1WI FSE PI
Fov (mm)	240×240	240×240	240×240	240×240	240×240	240×240
TR/TE (ms)	4800/120	4800/120	662/8.16	662/8.16	789/8.12	789/8.12
Matrix	384×269	384×269	384×307	384×307	384×307	384×307
ETL	28	28	2	2	2	2
Bandwidth (Hz)	260	260	280	280	250	250
Average	1	1	1	1	1	1
Number slices	40	40	40	40	40	40
Spatial resolution	0.89×0.63×5	0.89×0.63×5	0.78×0.63×5	0.78×0.63×5	0.78×0.63×5	0.78×0.63×5
Acquisition	ACS	PI	ACS	PI	ACS	PI
Accelerating factors	2.25	2	2.25	2	2.25	2
Sequence acquisition time (s)	59	82	93	140	75	90

Fig.1 Workflow of image preprocessing and consistency analysis between the two accelerated MRI sequences.

Fig.2 Scatter plot of intraclass correlation coefficients (ICCs) for all radiomic features extracted from T1WI, T2WI, and contrast-enhanced T1WI sequences, categorized by consistency levels. The red dashed line represents the ICC threshold of 0.75.

Fig.3 Boxplots of ICCs for different categories of radiomic features extracted from T1WI, T2WI, and contrast-enhanced T1WI (CE-T1WI) sequences. Each boxplot represents the distribution of ICCs within a specific feature category. The red dashed line indicates the reproducibility threshold at ICC=0.75, above which features are considered to have good agreement.

Fig.4 ROC curves of radiomics models constructed using PI- and ACS-accelerated images. The plots show the ROC curves of models built using all extracted features (A) and models built using only features with ICC>0.75 (B). The orange curve represents the model based on PI-accelerated images, and the blue curve represents the model based on ACS-accelerated images.

Tab.3 Predictive performance of different acceleration sequences

Feature type	Method	Accuracy	AUC	F1-Score	Sensitivity	Specificity	P
All Features	PI	0.891	0.889	0.931	0.940	0.714	0.991
All Features	ACS	0.891	0.902	0.931	0.940	0.714	0.991
Features ICC>0.75	PI	0.891	0.916	0.931	0.940	0.714	0.998
Features ICC>0.75	ACS	0.906	0.916	0.939	0.920	0.857	0.998

Fig.5 SHAP summary plots of feature contributions in random forest models. The plots show feature contributions for models constructed using features extracted from PI-accelerated images (A) and ACS-accelerated images (B).

References 47

[1]	Chen YP, Chan ATC, Le QT, et al. Nasopharyngeal carcinoma[J]. Lancet, 2019, 394(10192): 64-80. doi：10.1016/s0140-6736(19)30956-0
[2]	Ferlay J, Colombet M, Soerjomataram I, et al. Cancer statistics for the year 2020: an overview[J]. Int J Cancer, 2021, 149(4): 778-89. doi：10.1002/ijc.33588
[3]	Liao XB, Mao YP, Liu LZ, et al. How does magnetic resonance imaging influence staging according to AJCC staging system for nasopharyngeal carcinoma compared with computed tomography [J]? Int J Radiat Oncol Biol Phys, 2008, 72(5): 1368-77. doi：10.1016/j.ijrobp.2008.03.017
[4]	Wang F, Chen G, Zhang Z, et al. The Chinese society of clinical oncology (CSCO): clinical guidelines for the diagnosis and treatment of colorectal cancer, 2024 update[J]. Cancer Commun: Lond, 2025, 45(3): 332-79. doi：10.1002/cac2.12639
[5]	Deshmane A, Gulani V, Griswold MA, et al. Parallel MR imaging[J]. J Magn Reson Imaging, 2012, 36(1): 55-72. doi：10.1002/jmri.23639
[6]	Lustig M, Donoho D, Pauly JM. Sparse MRI: The application of compressed sensing for rapid MR imaging[J]. Magn Reson Med, 2007, 58(6): 1182-95. doi：10.1002/mrm.21391
[7]	Griswold MA, Jakob PM, Heidemann RM, et al. Generalized autocalibrating partially parallel acquisitions (GRAPPA)[J]. Magn Reson Med, 2002, 47(6): 1202-10. doi：10.1002/mrm.10171
[8]	Pruessmann KP, Weiger M, Scheidegger MB, et al. SENSE: Sensitivity encoding for fast MRI[J], Magn Reson Med， 1999;42(5): 952-62. doi：10.1002/(sici)1522-2594(199911)42:5<952::aid-mrm16>3.3.co;2-j
[9]	Fransen SJ, Roest C, Simonis FFJ, et al. The scientific evidence of commercial AI products for MRI acceleration: a systematic review[J]. Eur Radiol, 2025, 35(8): 4736-46. doi：10.1007/s00330-025-11423-5
[10]	Lin DJ, Johnson PM, Knoll F, et al. Artificial intelligence for MR image reconstruction: an overview for clinicians[J]. J Magn Reson Imaging, 2021, 53(4): 1015-28. doi：10.1002/jmri.27078
[11]	Ghodrati V, Shao J, Bydder M, et al. MR image reconstruction using deep learning: evaluation of network structure and loss functions[J]. Quant Imaging Med Surg, 2019, 9(9): 1516-27. doi：10.21037/qims.2019.08.10
[12]	Sheng RF, Zheng LY, Jin KP, et al. Single-breath-hold T2WI liver MRI with deep learning-based reconstruction: a clinical feasibility study in comparison to conventional multi-breath-hold T2WI liver MRI[J]. Magn Reson Imag, 2021, 81: 75-81. doi：10.1016/j.mri.2021.06.014
[13]	Liu HB, Deng DL, Zeng WL, et al. AI-assisted compressed sensing and parallel imaging sequences for MRI of patients with nasopharyngeal carcinoma: comparison of their capabilities in terms of examination time and image quality[J]. Eur Radiol, 2023, 33(11): 7686-96. doi：10.1007/s00330-023-09742-6
[14]	Wang QZ, Zhao WL, Xing XY, et al. Feasibility of AI-assisted compressed sensing protocols in knee MR imaging: a prospective multi-reader study[J]. Eur Radiol, 2023, 33(12): 8585-96. doi：10.1007/s00330-023-09823-6
[15]	王思敏, 李彦, 严福华, 等. 不同医用磁共振成像设备在中枢神经系统的临床图像质量评价与比较研究[J]. 磁共振成像, 2019(2): 83-8.
[16]	Ni M, He M, Yang YX, et al. Application research of AI-assisted compressed sensing technology in MRI scanning of the knee joint: 3D-MRI perspective[J]. Eur Radiol, 2024, 34(5): 3046-58. doi：10.1007/s00330-023-10368-x
[17]	Zhao Q, Xu J, Yang YX, et al. AI-assisted accelerated MRI of the ankle: clinical practice assessment[J]. Eur Radiol Exp, 2023, 7(1): 62. doi：10.1186/s41747-023-00374-5
[18]	Lambin P, Leijenaar RTH, Deist TM, et al. Radiomics: the bridge between medical imaging and personalized medicine[J]. Nat Rev Clin Oncol, 2017, 14(12): 749-62. doi：10.1038/nrclinonc.2017.141
[19]	Liu Z, Wang S, Dong D, et al. The applications of radiomics in precision diagnosis and treatment of oncology: opportunities and challenges[J]. Theranostics, 2019, 9(5): 1303-22. doi：10.7150/thno.30309
[20]	Kim MJ, Choi Y, Sung YE, et al. Early risk-assessment of patients with nasopharyngeal carcinoma: the added prognostic value of MR-based radiomics[J]. Transl Oncol, 2021, 14(10): 101180. doi：10.1016/j.tranon.2021.101180
[21]	Zhang L, Dong D, Li HL, et al. Development and validation of a magnetic resonance imaging-based model for the prediction of distant metastasis before initial treatment of nasopharyngeal carcinoma: a retrospective cohort study[J]. EBioMedicine, 2019, 40: 327-35. doi：10.1016/j.ebiom.2019.01.013
[22]	Zhang B, Tian J, Dong D, et al. Radiomics features of multiparametric MRI as novel prognostic factors in advanced nasopharyngeal carcinoma[J]. Clin Cancer Res, 2017, 23(15): 4259-69. doi：10.1158/1078-0432.ccr-16-2910
[23]	Zhang LL, Huang MY, Li Y, et al. Pretreatment MRI radiomics analysis allows for reliable prediction of local recurrence in non-metastatic T4 nasopharyngeal carcinoma[J]. EBioMedicine, 2019, 42: 270-80. doi：10.1016/j.ebiom.2019.03.050
[24]	张璐. 基于磁共振影像组学的鼻咽癌远处转移风险评估模型构建及验证[D]. 南方医科大学, 2019.
[25]	Teng X, Wang Y, Nicol AJ, et al. Enhancing the clinical utility of radiomics: addressing the challenges of repeatability and repro-ducibility in CT and MRI[J]. Diagnostics: Basel, 2024, 14(16): 1835. doi：10.3390/diagnostics14161835
[26]	Li S, Deng YQ, Zhu ZL, et al. A comprehensive review on radiomics and deep learning for nasopharyngeal carcinoma imaging[J]. Diagnostics: Basel, 2021, 11(9): 1523. doi：10.3390/diagnostics11091523
[27]	Joo L, Jung SC, Lee H, et al. Stability of MRI radiomic features according to various imaging parameters in fast scanned T2-FLAIR for acute ischemic stroke patients[J]. Sci Rep, 2021, 11(1): 17143. doi：10.1038/s41598-021-96621-z
[28]	Rai R, Barton MB, Chlap P, et al. Repeatability and reproducibility of magnetic resonance imaging-based radiomic features in rectal cancer[J]. J Med Imaging: Bellingham, 2022, 9(4): 044005. doi：10.1117/1.jmi.9.4.044005
[29]	Klontzas ME. Radiomics feature reproducibility: The elephant in the room[J]. Eur J Radiol, 2024, 175: 111430. doi：10.1016/j.ejrad.2024.111430
[30]	Reschke P, Koch V, Gruenewald LD, et al. Deep learning-accelerated prostate MRI: improving speed, accuracy, and sustainability[J]. Acad Radiol, 2025, 14: S1076-6332(25)00571-9. doi：10.1016/j.acra.2025.06.022
[31]	Schwier M, van Griethuysen J, Vangel MG, et al. Repeatability of multiparametric prostate MRI radiomics features[J]. Sci Rep, 2019, 9(1): 9441. doi：10.1038/s41598-019-45766-z
[32]	Zwanenburg A, Vallières M, Abdalah MA, et al. The image biomarker standardization initiative: standardized quantitative radiomics for high-throughput image-based phenotyping[J]. Radiology, 2020, 295(2): 328-38. doi：10.1148/radiol.2020191145
[33]	Granzier RWY, Ibrahim A, Primakov S, et al. Test-retest data for the assessment of breast MRI radiomic feature repeatability[J]. J Magn Reson Imaging, 2022, 56(2): 592-604. doi：10.1002/jmri.28027
[34]	Lu H, Parra NA, Qi J, et al. Repeatability of quantitative imaging features in prostate magnetic resonance imaging[J]. Front Oncol, 2020, 10: 551. doi：10.3389/fonc.2020.00551
[35]	Scalco E, Belfatto A, Mastropietro A, et al. T2w-MRI signal normalization affects radiomics features reproducibility[J]. Med Phys, 2020, 47(4): 1680-91. doi：10.1002/mp.14038
[36]	Koo TK, Li MY. A guideline of selecting and reporting intraclass correlation coefficients for reliability research[J]. J Chiropr Med, 2016, 15(2): 155-63. doi：10.1016/j.jcm.2016.02.012
[37]	Bi Y, Xiang DX, Ge ZY, et al. An interpretable prediction model for identifying N7-methylguanosine sites based on XGBoost and SHAP[J]. Mol Ther Nucleic Acids, 2020, 22: 362-72. doi：10.1016/j.omtn.2020.08.022
[38]	Shukla-Dave A, Obuchowski NA, Chenevert TL, et al. Quantitative imaging biomarkers alliance (QIBA) recommendations for improved precision of DWI and DCE-MRI derived biomarkers in multicenter oncology trials[J]. J Magn Reson Imaging, 2019, 49(7): e101-21. doi：10.1002/jmri.26805
[39]	Salluzzi M, McCreary CR, Gobbi DG, et al. Short-term repeatability and long-term reproducibility of quantitative MR imaging biomarkers in a single centre longitudinal study[J]. NeuroImage, 2022, 260: 119488. doi：10.1016/j.neuroimage.2022.119488
[40]	Ye S, Lim JY, Huang W. Statistical considerations for repeatability and reproducibility of quantitative imaging biomarkers[J]. BJR Open, 2022, 4(1): 20210083. doi：10.1259/bjro.20210083
[41]	Orlhac F, Lecler A, Savatovski J, et al. How can we combat multicenter variability in MR radiomics? Validation of a correction procedure[J]. Eur Radiol, 2021, 31(4): 2272-80. doi：10.1007/s00330-020-07284-9
[42]	Carré A, Battistella E, Niyoteka S, et al. AutoComBat: a generic method for harmonizing MRI-based radiomic features[J]. Sci Rep, 2022, 12(1): 12762. doi：10.1038/s41598-022-16609-1
[43]	Zhao W, Hu Z, Kazerooni AF, et al. Physics-informed discretization for reproducible and robust radiomic feature extraction using quantitative MRI[J]. Invest Radiol, 2024, 59(5): 359-71. doi：10.1097/rli.0000000000001026
[44]	胡丽霞, 朱进, 周阳泱, 等. 鼻咽癌动态增强MRI参数与临床病理特征的关系[J]. 中国医学影像技术, 2014, 30(4): 514-7．
[45]	Carbonell G, Kennedy P, Bane O, et al. Precision of MRI radiomics features in the liver and hepatocellular carcinoma[J]. Eur Radiol, 2022, 32(3): 2030-40. doi：10.1007/s00330-021-08282-1
[46]	Perrin T, Midya A, Yamashita R, et al. Short-term reproducibility of radiomic features in liver parenchyma and liver malignancies on contrast-enhanced CT imaging[J]. Abdom Radiol, 2018, 43(12): 3271-8. doi：10.1007/s00261-018-1600-6
[47]	Ibrahim A, Guha S, Lu L, et al. The reproducibility and predictivity of radiomic features extracted from dynamic contrast-enhanced computed tomography of hepatocellular carcinoma[J]. PLoS One, 2024, 19(9): e0310486. doi：10.1371/journal.pone.0310486