EUVIMEDEUVIMED entdecken
Crossref

Deep learning-based exploration of radiomics data for modeling and predicting the therapeutic efficacy of methotrexate in treating juvenile idiopathic arthritis

Haojie Qiu, Ping Xie, Wenjing Gao, Wen He, Yousi Miao, Zhuolin Yang, Jinlian Yang, Yuanyi Wei, Wenqing Wang, Yiting Ou, Wanhuan Feng, Ping Zeng und 6 weitere

Translational Pediatrics · 2026-11 · 15 (11) · 462-462

DOI: 10.21037/tp-2026-0587 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Translational Pediatrics, 2026-11. Verfasst von Haojie Qiu und 17 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Background

There is significant individual variability in the efficacy of methotrexate (MTX), making its precise application in clinical practice challenging. Currently, there are no clear radiomic biomarkers available to predict the efficacy of MTX treatment in patients with juvenile idiopathic arthritis (JIA). Based on the magnetic resonance imaging (MRI) characteristics of juvenile idiopathic arthritis (JIA) patients before and after methotrexate (MTX) treatment, deep learning was used to mine the imaging omics information in order to develop a non-invasive, convenient and highly applicable imaging omics prediction model for early prediction of MTX treatment for JIA.

Methods

MRI images of 103 JIA patients treated with oral MTX monotherapy at Guangzhou Women and Children's Medical Center between 2008 and 2019 were retrospectively collected. Firstly, a clinician with more than 5 years of work experience used LabelMe (version 3.16.7) to manually segment the inflammatory lesions from two-dimensional (2D) MRI images. Next, six different segmentation algorithms (UNet, AttU_Net, UnET ++, Deeplabv3plus_resnet101, AttR2U_Net and ResUNet) were used to construct the segmentation model by Pytorch (Version:1.11). Then four different neural network algorithms (ResNet_152, DensNet_161, VGG_16 and ConvNext) were used to construct therapeutic efficacy classification models, and finally the optimal models were compared. The sample composition of model training set and test set was 93:10.

Results

Based on MRI data, we used deep learning and imaging omics methods to develop a segmentation model for JIA knee lesions and a prediction model for MTX treatment for JIA. Among them, the segmentation model developed by ResUNet performed the best, with a Dice of 0.7930, intersection over union (IoU) of 0.6570, lesion pixel accuracy (ACC) of 0.7973, Recall of 0.7886, Precision of 0.7973, and F1-score of 0.7930. Meanwhile, the classification model developed using ConvNext achieved the best performance, with an area under the curve (AUC) of 0.6137, sensitivity of 0.5684, specificity of 0.5111, precision of 0.4500, and accuracy of 0.5347.

Conclusions

In this study, deep learning was used to mine patients' MRI image data, and models were constructed for automatic segmentation of knee inflammation in children with JIA and prediction of MTX efficacy. The model achieved good performance in detecting and segmentating JIA knee inflammation, and was able to predict whether MTX is effective. The model has the advantages of being non-invasive and convenient, which can be applied by clinicians and pharmacists to enhance clinically assisted decision making.

Zitieren

APA

Qiu, H., Xie, P., Gao, W., He, W., Miao, Y., Yang, Z., Yang, J., Wei, Y., Wang, W., Ou, Y., Feng, W., Zeng, P., Zhou, J., Li, J., Hu, B., Wen, Z., Liu, H., & Mo, X. (2026). Deep learning-based exploration of radiomics data for modeling and predicting the therapeutic efficacy of methotrexate in treating juvenile idiopathic arthritis. Translational Pediatrics, 15(11), 462-462. https://doi.org/10.21037/tp-2026-0587

Vancouver

Qiu H, Xie P, Gao W, He W, Miao Y, Yang Z, et al. Deep learning-based exploration of radiomics data for modeling and predicting the therapeutic efficacy of methotrexate in treating juvenile idiopathic arthritis. Translational Pediatrics. 2026;15(11):462-462. doi:10.21037/tp-2026-0587

BibTeX
@article{Qiu2026,
  author = {Qiu, Haojie and Xie, Ping and Gao, Wenjing and He, Wen and Miao, Yousi and Yang, Zhuolin and Yang, Jinlian and Wei, Yuanyi and Wang, Wenqing and Ou, Yiting and Feng, Wanhuan and Zeng, Ping and Zhou, Juan and Li, Jiali and Hu, Bimin and Wen, Zehuang and Liu, Hongsheng and Mo, Xiaolan},
  title = {Deep learning-based exploration of radiomics data for modeling and predicting the therapeutic efficacy of methotrexate in treating juvenile idiopathic arthritis},
  journal = {Translational Pediatrics},
  year = {2026},
  volume = {15},
  number = {11},
  pages = {462--462},
  doi = {10.21037/tp-2026-0587}
}
Kostenlos starten →