Page 41 - ARNM-3-2

Advances in Radiotherapy
& Nuclear Medicine Radiomics for gastric cancer

than 80%, with the best AUC of 0.908 in the validation leading to inconsistencies in feature extraction and model
set. 36-48 Since PNI and LVI have been identified as important reproducibility. One limitation of these studies is that
prognostic factors, the AUCs of their predictive models they were conducted at a single center with a single image
have improved significantly in recent years (LVI = 0.777 – modality, which limits the stability and practicability of
0.921, PNI = 0.482 – 0.820). 50-54 the constructed models. In future studies, the sharing
88
The prediction of prognosis is of great significance of multicenter data should be conducted to improve the
for patients’ risk management and treatment. Studies effectiveness and practicability of the radiomic models.
83
have confirmed the value of surgery, chemotherapy, and Another direction is using multi-modal hybrid images
other therapeutic methods in the management of GC. for radiomic studies, which inherit the advantages of
84
In addition, the feasibility of radiomics for predicting each single image and contain more types of information
89
survival or recurrence has been confirmed and verified, to be developed. In addition, while deep learning has
with C-index values ranging from 0.710 to 0.820. 58-73 shown its potential in automating image segmentation,
However, radiotherapy is also one of the main methods for its integration with radiomics is still in its early stages,
the treatment of malignant tumors, and there are relatively and challenges such as model interpretability and data
few radiomic studies on GC patients who have undergone requirements remain unanswered. Finally, the clinical
85
radiotherapy. Therefore, predicting the efficacy of translation of radiomics is hindered by issues such
radiotherapy or chemoradiotherapy in advance using as data standardization, regulatory approval, and the
radiomics holds great clinical significance for GC patients. need for multicenter validation. Integrating nomogram
and radiomic features with physiological information,
Radiomics has shown great potential in predicting pathological indicators, and gene expression may show
personalized NAC responses in patients with advanced GC. promising ability to increase the accuracy and stability of
Radiomic models have demonstrated a stable performance the prediction models in future studies.
in predicting NAC response, with accuracies ranging from
0.722 to 0.763. 62,64 The integration of machine learning 5. Conclusion
further improved the prediction accuracy, as demonstrated
in a study by Xu et al., where the AUC increased from Radiomics can deeply mine medical images to reveal
65
0.750 to 0.889 across validation cohorts. The combination tumor heterogeneity, quantify tumor information, reflect
of radiomic and clinical features also enhanced survival potential molecular biological changes, and improve
prediction, with Chen et al. ’s model achieving C-indices individualized treatment choices. With the continuous
67
of 0.810 for DFS and 0.710 for OS. improvement and development of radiomics, it has great
potential in the diagnosis, post-treatment follow-up, and
CT is the main detection method for GC; thus, most treatment decision-making for patients with GC.
radiomic studies are based on CT images. Although
PET-CT can analyze the functional and metabolic Acknowledgments
information of the lesion qualitatively and quantitatively
with clear anatomy, the number of cases available for None.
radiomic studies is limited. Similarly, due to the limited Funding
86
number of cases detected by MRI, there have been few
radiomic studies on GC using MRI in recent years. This study is financially supported by the Joint Laboratory
Therefore, radiomic studies based on PET-CT and MRI of Basic Medicine and Intelligent Medicine.
for GC patients are of great potential. Another future
direction may focus on endoscopic ultrasound, which has Conflict of interest
advantages in the diagnosis of submucosal tumors and Congying Xie is an Editorial Board Member of this
prediction of the depth of tumor invasion. 87 journal but was not in any way involved in the editorial
Radiomic studies in GC have several key limitations. and peer-review process conducted for this paper, directly
First, there is a lack of standardization in the delineation or indirectly. Separately, other authors declared that they
of regions of interest, with most studies relying on manual have no known competing financial interests or personal
or semi-automatic segmentation methods, which are time- relationships that could have influenced the work reported
consuming and prone to subjective bias. The irregular in this paper.
shape of gastric tumors further complicates accurate Author contributions
boundary detection, affecting model performance. Second,
imaging acquisition protocols, reconstruction algorithms, Conceptualization: Xiance Jin
and preprocessing methods vary across institutions, Investigation: Qiao Zheng, Haoze Zheng

Volume 3 Issue 2 (2025) 33 doi: 10.36922/arnm.8350