Page 34 - ARNM-3-2

Advances in Radiotherapy
& Nuclear Medicine Radiomics for gastric cancer

algorithm (n = 4). These retrospective studies consist of least absolute shrinkage, selection operator, and principal
39 (75%) single-center studies and 13 (25%) multicenter component analysis, are applied to reduce dimensionality
studies with sample sizes ranging from 47 to 1,778. There and retain the most relevant features. Finally, predictive
were 44 studies of CT radiomics, six studies of PET-CT, models are developed using machine learning algorithms
and a few studies of MRI. such as support vector machines or random forests, and the
performance is evaluated using metrics such as accuracy
3. Research directions of radiomics and sensitivity.

3.1. Radiomic workflow 3.2. Diagnosis of GC
In radiomics, the initial steps involve acquiring high- The diagnosis and identification of GC are the first step in
quality medical images from modalities such as ultrasound, treatment decision-making. Histological diagnosis with
CT, or MRI by following standardized protocols for endoscopic biopsy is usually applied and considered the
consistency. After obtaining the images, the region of gold standard for GC in the pre-operative diagnosis stage as
interest is identified and delineated manually or using it is hard to differentiate GC from other gastric tumors in the
segmentation algorithms. Once defined, quantitative imaging, such as gastric stromal tumors and lymphoma. 18,19
features such as mean, variance, volume, surface area, and However, radiomics based on CT and MRI images provide
texture, such as gray-level co-occurrence matrix (GLCM) a non-invasive method for the differentiation of GC, as
and gray-level run-length matrix are extracted using tools shown in Table 1. Ma et al. evaluated the value of portal
20
like PyRadiomics. Feature selection methods, such as venous CT-based radiomic signature in differentiating

Table 1. Summary of studies on radiomics to identify gastric cancer
Authors Year Image Purpose Sample Features Results
Ma et al. 20 2017 CT Differentiation of Borrmann 40 (Single center) Radiomics The combined model showed a diagnostic accuracy
type IV GC from primary + clinical of 87.14% (AUC=0.903)
gastric lymphoma
Sun et al. 21 2020 PET-CT Distinguish between GC and 79 (Single center) Radiomics The model had an AUC of 0.770 in distinguishing
gastric lymphoma between the gastric low-grade lymphoma and GC
Sun et al. 22 2019 CT Distinguish between GC and 100 (Single center) Radiomics The combined model yielded the highest AUC value
gastric stromal tumor + clinical (0.903), specificity (93.33%), and accuracy (86.00%)
among the three models
Wang et al. 23 2021 CT Differentiating gastric 63 (Single center) Radiomics The nomogram incorporated with the radiomic
neuroendocrine carcinoma + clinical signature, tumor margin, and LNM showed an AUC of
from gastric ADC 0.821 in the primary cohort and 0.809 in the validation
cohort
Zhang et al. 26 2017 MRI Predicting histological grade 78 (Single center) Radiomics The AUC value of the maximum frequency
of GC parameter in distinguishing the differentiation
degree of GC was the largest, which was 0.675
Li et al. 27 2019 CT To predict the adverse 554 (Single center) Radiomics The developed calculation method showed good
histopathological status of + clinical performance in predicting GC AHS and clinical
GC disease
Wang et al. 30 2020 CT Lauren diffuse type and 539 (Single center) Radiomics The radiomic nomogram incorporating the
intestinal type of GC were + clinical combined radiomic signature, age, T stage, and
distinguished before surgery N stage outperformed the other models with a
training AUC of 0.745 and a validation AUC of
0.758
Wang et al. 31 2020 CT To investigate the role 187 (Single center) Radiomics The nomogram yielded excellent performance for
of CT radiomics for the distinguishing intestinal-type ADCs in training and
pre-operative distinction of test sets, with AUCs of 0.928 and 0.904, respectively
intestinal-type gastric ADCs
Chen et al. 32 2022 CT To distinguish between diffuse 693 (Multi-center) Radiomics The radiomic nomogram showed good
type and signet ring cell + clinical discrimination, with AUC of 0.905, 0.845, and
carcinoma GC before surgery 0.918 in the training, internal, and external
validation cohorts, respectively
Abbreviations: ADC: Adenocarcinoma; AHS: Adverse histopathological status; AUC: Area under the curve; CT: Computed tomography; GC: Gastric
cancer; MRI: Magnetic resonance imaging; PET-CT: Positron emission tomography-computed tomography.

Volume 3 Issue 2 (2025) 26 doi: 10.36922/arnm.8350