Share：
Share this content in WeChat
X
[Chinese] [PDF] 10 1
• Clinical Article •
Comparison of MRI radiomics models combined with clinical features for predicting treatment efficacy in adenomyosis after high intensity focused ultrasound using different machine learning algorithm
ZHU Zhijun  HUANG Xiaohua  ZHOU Mengni  LUO Jingxian  LI Yanting  LIU Ziyan  LIU Ziyi 

DOI:10.12015/issn.1674-8034.2025.12.020.


[Abstract] Objective To compare the efficacy of MRI radiomics models combined with clinical features, constructed using different machine learning (ML) algorithms, for predicting treatment outcomes in adenomyosis patients after high-intensity focused ultrasound (HIFU).Materials and Methods Imaging and clinical data from 169 adenomyosis patients who underwent HIFU treatment between September 2021 and May 2024 and met inclusion/exclusion criteria were retrospectively collected. Postoperative non-perfused volume (NPV) was assessed via MRI. Patients were stratified into a significant-efficacy group [NPV ratio (NPVR) ≥ 50%, n = 76] and a non-significant-efficacy group (NPVR < 50%, n = 93) using the threshold NPVR = 50% (NPVR = NPV / total lesion volume). Lesions were segmented using 3D Slicer software for feature extraction. Eight ML algorithms were used to build models: decision tree (DT), Gaussian process (GP), logistic regression (LR), partial least squares discriminant analysis (PLSDA), quadratic discriminant analysis (QDA), random forest (RF), stochastic gradient descent (SGD), and support vector machine (SVM). Model performance was evaluated using receiver operating characteristic (ROC) curves, with calculation of the area under the curve (AUC), sensitivity, specificity, accuracy, precision, and F1-score. DeLong test compared inter-model differences (statistical significance: P < 0.05).Results Radiomics-clinical models based on DT, GP, LR, PLSDA, QDA, RF, SGD, and SVM algorithms were constructed. Training set AUCs were: 0.865 (95% CI: 0.806 to 0.924), 0.713 (95% CI: 0.619 to 0.807), 0.666 (95% CI: 0.567 to 0.764), 0.669 (95% CI: 0.571 to 0.767), 0.649 (95% CI: 0.550 to 0.749), 0.796 (95% CI: 0.717 to 0.876), 0.425 (95% CI: 0.341 to 0.508), and 0.666 (95% CI: 0.568 to 0.764), respectively. Test set AUCs were: 0.788 (95% CI: 0.669 to 0.907), 0.738 (95% CI: 0.601 to 0.874), 0.719 (95% CI: 0.578 to 0.860), 0.730 (95% CI: 0.592 to 0.868), 0.738 (95% CI: 0.600 to 0.875), 0.731 (95% CI: 0.587 to 0.876), 0.332 (95% CI: 0.221 to 0.444), and 0.719 (95% CI: 0.575 to 0.863), respectively. The DT model achieved the highest AUC, specificity, precision, and accuracy in the test set, and the highest AUC and F1-score in the training set. SGD and PLSDA models performed poorly in both sets.Conclusions MRI radiomics-clinical models built using six ML algorithms (DT, GP, LR, QDA, RF, SVM) demonstrated good predictive performance for post-HIFU efficacy in adenomyosis. The DT model exhibited optimal performance and is recommended as the preferred method for outcome prediction, assisting clinicians in developing personalized treatment plans and management strategies.
[Keywords] adenomyosis；high-intensity focused ultrasound；magnetic resonance imaging；radiomics；machine learning

ZHU Zhijun   HUANG Xiaohua*   ZHOU Mengni   LUO Jingxian   LI Yanting   LIU Ziyan   LIU Ziyi  

Department of Radiology, Affiliated Hispital of North Sichuan Medical College, Nanchong 637000, China

Corresponding author: HUANG X H, E-mail: 15082797553@163.com

Conflicts of interest   None.

Received  2025-07-28
Accepted  2025-12-06
DOI: 10.12015/issn.1674-8034.2025.12.020
DOI:10.12015/issn.1674-8034.2025.12.020.

[1]
KHO K A, CHEN J S, HALVORSON L M. Diagnosis, evaluation, and treatment of adenomyosis[J]. JAMA, 2021, 326(2): 177-178. DOI: 10.1001/jama.2020.26436.
[2]
YOUNES G, TULANDI T. Effects of adenomyosis on in vitro fertilization treatment outcomes: a meta-analysis[J/OL]. Fertil Steril, 2017, 108(3): 483-490.e3 [2025-07-09]. https://pubmed.ncbi.nlm.nih.gov/28865548/. DOI: 10.1016/j.fertnstert.2017.06.025.
[3]
JIN W T, WANG S J, WANG T P, et al. Multi-machine learning model based on habitat subregions for outcome prediction in adenomyosis treated by uterine artery embolization[J]. Acad Radiol, 2024, 31(12): 4985-4995. DOI: 10.1016/j.acra.2024.05.037.
[4]
DASON E S, CHAN C, SOBEL M. Diagnosis and treatment of adenomyosis[J/OL]. CMAJ, 2021, 193(7): E242 [2025-07-09]. https://pubmed.ncbi.nlm.nih.gov/33593948/. DOI: 10.1503/cmaj.201607.
[5]
ZHANG L, RAO F W, SETZEN R. High intensity focused ultrasound for the treatment of adenomyosis: selection criteria, efficacy, safety and fertility[J]. Acta Obstet Gynecol Scand, 2017, 96(6): 707-714. DOI: 10.1111/aogs.13159.
[6]
BURLA L, SARTORETTI E, MANNIL M, et al. MRI-based radiomics as a promising noninvasive diagnostic technique for adenomyosis[J/OL]. J Clin Med, 2024, 13(8): 2344 [2025-07-09] https://pubmed.ncbi.nlm.nih.gov/38673617/.. DOI: 10.3390/jcm13082344.
[7]
CHAPRON C, VANNUCCINI S, SANTULLI P, et al. Diagnosing adenomyosis: an integrated clinical and imaging approach[J]. Hum Reprod Update, 2020, 26(3): 392-411. DOI: 10.1093/humupd/dmz049.
[8]
LI J Q, WANG W, LIAO L J, et al. Analysis of the nonperfused volume ratio of adenomyosis from MRI images based on fewshot learning[J/OL]. Phys Med Biol, 2021, 66(4): 045019 [2025-07-09]. https://pubmed.ncbi.nlm.nih.gov/33361557/. DOI: 10.1088/1361-6560/abd66b.
[9]
KESERCI B, DUC N M. Magnetic resonance imaging features influencing high-intensity focused ultrasound ablation of adenomyosis with a nonperfused volume ratio of ≥90% as a measure of clinical treatment success: retrospective multivariate analysis[J]. Int J Hyperthermia, 2018, 35(1): 626-636. DOI: 10.1080/02656736.2018.1516301.
[10]
FAN T Y, ZHANG L, CHEN W Z, et al. Feasibility of MRI-guided high intensity focused ultrasound treatment for adenomyosis[J]. Eur J Radiol, 2012, 81(11): 3624-3630. DOI: 10.1016/j.ejrad.2011.05.036.
[11]
YU J W, YANG M J, JIANG L, et al. Factors influencing USgHIFU ablation for adenomyosis with NPVR ≥ 50[J/OL]. Int J Hyperthermia, 2023, 40(1): 2211753 [2025-07-09]. https://pubmed.ncbi.nlm.nih.gov/37211351/. DOI: 10.1080/02656736.2023.2211753.
[12]
YING J, JING X, GAO F, et al. Prediction of ablation rate for high-intensity focused ultrasound therapy of adenomyosis in MR images based on multi-model fusion[J]. J Imaging Inform Med, 2024, 37(4): 1579-1590. DOI: 10.1007/s10278-024-01063-4.
[13]
LI C W, HE Z M, LV F J, et al. An interpretable MRI-based radiomics model predicting the prognosis of high-intensity focused ultrasound ablation of uterine fibroids[J/OL]. Insights Imaging, 2023, 14(1): 129 [2025-07-16]. https://pubmed.ncbi.nlm.nih.gov/37466728/. DOI: 10.1186/s13244-023-01445-2.
[14]
ZHENG Y N, CHEN L P, LIU M Q, et al. Prediction of clinical outcome for high-intensity focused ultrasound ablation of uterine leiomyomas using multiparametric MRI radiomics-based machine leaning model[J/OL]. Front Oncol, 2021, 11: 618604 [2025-07-16]. https://pubmed.ncbi.nlm.nih.gov/34567999/. DOI: 10.3389/fonc.2021.618604.
[15]
GONG C M, SETZEN R, LIU Z Q, et al. High intensity focused ultrasound treatment of adenomyosis: The relationship between the features of magnetic resonance imaging on T2 weighted images and the therapeutic efficacy[J]. Eur J Radiol, 2017, 89: 117-122. DOI: 10.1016/j.ejrad.2017.02.001.
[16]
LIU Z Y, LIU Z Y, WAN X Y, et al. Prediction of clinical outcome for high-intensity focused ultrasound ablation of adenomyosis based on non-enhanced MRI radiomics[J/OL]. Int J Hyperthermia, 2025, 42(1): 2468766 [2025-07-13]. https://pubmed.ncbi.nlm.nih.gov/39988330/. DOI: 10.1080/02656736.2025.2468766.
[17]
SUN M M, DENG X Y, XING H, et al. A dual-sequence MRI-based radiomics model for predicting high-intensity focused ultrasound ablation efficacy in adenomyosis treatment[J]. Int J Womens Health, 2025, 17: 1321-1332. DOI: 10.2147/IJWH.S512216.
[18]
LIU J F, GUO W, ZENG P E, et al. Vertebral MRI-based radiomics model to differentiate multiple myeloma from metastases: influence of features number on logistic regression model performance[J]. Eur Radiol, 2022, 32(1): 572-581. DOI: 10.1007/s00330-021-08150-y.
[19]
PEDUZZI P, CONCATO J, FEINSTEIN A R, et al. Importance of events per independent variable in proportional hazards regression analysis. II. Accuracy and precision of regression estimates[J]. J Clin Epidemiol, 1995, 48(12): 1503-1510. DOI: 10.1016/0895-4356(95)00048-8.
[20]
SI M, LV F J, TANG M M, et al. Non-contrast enhanced MRI for efficiency evaluation of high-intensity focused ultrasound in adenomyosis ablation[J/OL]. Int J Hyperthermia, 2024, 41(1): 2295813 [2025-07-09]. https://pubmed.ncbi.nlm.nih.gov/38234000/. DOI: 10.1080/02656736.2023.2295813.
[21]
YU J W, JIANG L, SU X Y, et al. Comparison efficacy of ultrasound-guided HIFU for adenomyosis-associated dysmenorrhea with different signal intensity on T2-weighted MR imaging[J]. J Obstet Gynaecol Res, 2023, 49(4): 1189-1197. DOI: 10.1111/jog.15567.
[22]
LAMBIN P, LEIJENAAR R T H, DEIST T M, et al. Radiomics: the bridge between medical imaging and personalized medicine[J]. Nat Rev Clin Oncol, 2017, 14(12): 749-762. DOI: 10.1038/nrclinonc.2017.141.
[23]
YU H H, JIANG L, PENG M Y, et al. Developing an interpretable model to predict hemorrhagic transformation risk in acute stroke using multiparameter MRI[J]. Chin J Magn Reson Imag, 2025, 16(4): 19-24. DOI: 10.12015/issn.1674-8034.2025.04.004.
[24]
XU Z C, ZHANG Y, LIU Q, et al. Differentiation of high-grade glioma and metastatic tumor based on MRI radiomics and semantic features[J]. Chin J Magn Reson Imag, 2024, 15(8): 103-109, 123. DOI: 10.12015/issn.1674-8034.2024.08.016.
[25]
XU G, CHEN P, LI Y L, et al. Study on the ability to grade the risk of cervical spondylotic myelopathy by using machine learning model based on MRI radiomics[J]. Chin J Magn Reson Imag, 2024, 15(4): 50-55, 77. DOI: 10.12015/issn.1674-8034.2024.04.009.
[26]
HANDELMAN G S, KOK H K, CHANDRA R V, et al. eDoctor: machine learning and the future of medicine[J]. J Intern Med, 2018, 284(6): 603-619. DOI: 10.1111/joim.12822.
[27]
YU W J, LIU A F, CHEN J X, et al. Advantages and application strategies of machine learning in diagnosis and treatment of lumbar disc herniation[J]. Chin J Tissue Eng Res, 2024, 28(9): 1426-1435. DOI: 10.12307/2023.904.
[28]
KHAN A, KHAN A, KHAN M M, et al. Cardiovascular and diabetes diseases classification using ensemble stacking classifiers with SVM as a Meta classifier[J/OL]. Diagnostics (Basel), 2022, 12(11): 2595 [2025-07-14]. https://pubmed.ncbi.nlm.nih.gov/36359438/. DOI: 10.3390/diagnostics12112595.
[29]
WANG L N, YANG Y, MIN R Q, et al. Accelerating deep neural network training with inconsistent stochastic gradient descent[J]. Neural Netw, 2017, 93: 219-229. DOI: 10.1016/j.neunet.2017.06.003.
[30]
MORTAZI A, CICEK V, KELES E, et al. Selecting the best optimizers for deep learning-based medical image segmentation[J/OL]. Front Radiol, 2023, 3: 1175473 [2025-07-25]. https://pubmed.ncbi.nlm.nih.gov/37810757/. DOI: 10.3389/fradi.2023.1175473.
[31]
RUIZ-PEREZ D, GUAN H, MADHIVANAN P, et al. So you think you can PLS-DA [J/OL]. BMC Bioinformatics, 2020, 21(suppl 1): 2 [2025-07-14]. https://pubmed.ncbi.nlm.nih.gov/33297937/. DOI: 10.1186/s12859-019-3310-7.
[32]
FENG F, HASABALLA A I, LONG T, et al. AI-Driven segmentation and morphogeometric profiling of epicardial adipose tissue in type 2 diabetes[J/OL]. Cardiovasc Diabetol, 2025, 24(1): 294 [2025-07-25]. https://pubmed.ncbi.nlm.nih.gov/40682091/. DOI: 10.1186/s12933-025-02829-y.
[33]
LI Z C, ZHANG J, SONG Y, et al. Utilization of radiomics to predict long-term outcome of magnetic resonance-guided focused ultrasound ablation therapy in adenomyosis[J]. Eur Radiol, 2021, 31(1): 392-402. DOI: 10.1007/s00330-020-07076-1.
[34]
VY V P T, CHIEN J C, IRAMA W, et al. Machine learning models using multiparametric MRI for preoperative risk stratification in endometrial cancer[J]. Am J Cancer Res, 2024, 14(11): 5400-5410. DOI: 10.62347/MALY3908.
[35]
WANG W Y, ZENG W W, YANG S. A stacked machine learning-based classification model for endometriosis and adenomyosis: a retrospective cohort study utilizing peripheral blood and coagulation markers[J/OL]. Front Digit Health, 2024, 6: 1463419 [2025-07-14]. https://pubmed.ncbi.nlm.nih.gov/39347446/. DOI: 10.3389/fdgth.2024.1463419.

PREV Radiomics analysis of multi-sequence MRI evaluate lymphovascular space invasion in endometrial carcinoma
NEXT Functional magnetic resonance imaging for evaluating the efficacy of high-intensity focused ultrasound in uterine fibroid treatment
  


LINK


Tel & Fax: +8610-67113815    E-mail: editor@cjmri.cn