European Radiology: Radiomics in preoperative grading of pancreatic cancer

Pancreatic ductal adenocarcinoma (PDAC) is the most common pancreatic malignancy with a 5-year survival rate of about 7%.

80-85% of people with pancreatic cancer are unresectable
at the time of diagnosis.

It is well known that the determining factor in the prognosis of PDAC is the histopathological subtype
.
Well-differentiated PDAC (grade 1) is associated with
long-term postoperative survival.
Poor differentiation is an independent prognostic factor that affects overall survival
.
Given the poor prognosis of PDAC and the high incidence of postoperative complications, further risk stratification is needed clinically to develop more effective treatment strategies
.
In routine practice, the degree of differentiation of lesions is determined after surgical treatment
.
In order to improve the survival prognosis and quality of life of patients after surgical treatment, accurate preoperative diagnosis is of great
significance.
Despite the high risk of postoperative complications of pancreatitis, the only way to determine PDAC grade before surgery is invasive biopsy, which includes ultrasound-guided endoscopic fine-needle biopsy (EUS-FNB) and ultrasound/computed tomography (CT)-guided percutaneous biopsy
.
Therefore, a safe and accessible preoperative method is needed to determine the degree
of differentiation of PDAC.

Currently, the only way to reliably diagnose PDAC is by means of imaging such as ultrasound, CT, MRI, or comprehensive imaging
.
At present, the sensitivity and specificity of CT in the diagnosis of pancreatic cancer have reached 89% and 90%,
respectively.
Dynamic CT images of tumors show tumor spread, vascularization of tumor tissue, and invasion
of main blood vessels.
However, these imaging data are mostly nonspecific at the microstructure and molecular level, and much genetic and prognostic information remains unrevealed
.

Radiogenomics is a technique for establishing relationships between the genotype and the visualization phenotype of a tumor, closely
related to texture analysis.
Radiogenomics provides a comprehensive quantitative assessment
of tumor phenotypes by extracting a large number of quantitative features from medical images.
Therefore, radiomics can be used to reveal cancer for differential diagnosis, surgical strategy, prognosis, response prediction, and follow-up
.

Recently, a study published in the journal European Radiology developed a diagnostic algorithm based on radiomics model for the graded prediction of PDAC, which provides technical support
for clinical preoperative non-invasive evaluation.

This study included 91 patients with histologically confirmed PDAC and subgrouped
according to tumor grade.
Two radiologists blinded histologically independently segmented lesions and performed quantitative texture analysis
on all enhanced images.
Calculate the density ratio of PDAC and unchanged pancreatic tissue, as well as relative tumor enhancement (RTE)
in arterial, portal vein, and delayed examination stages.
Principal component analysis is used in multivariate predictor analysis
.
The selection of predictors in the binary logic model is carried out in 2 stages: (1) using a one-factor logic model (the selection criterion is P < 0.
1); (2) Use regularization (LASSO regression after variable normalization).

Among the 62 textured features in the arterial, portal venous and delayed phases, there were significant differences in 4, 16 and 8, respectively (P < 0.
1).

AFTER SCREENING, THE FINAL DIAGNOSTIC MODEL INCLUDED RADIOMIC FEATURES
SUCH AS DISCRETIZED HU CRITERIA, DISCRETIZED HUQ3, GGLCM-RELATED, PORTAL CONTRAST AGENT-ENHANCED GLZLM LZLGE, AND CONVENTIONAL_HUQ3 IN THE DELAYED PHASE OF CT STUDIES.
On its basis, a diagnostic model was built, showing that the AUC of ≥ grade 2 was 0.
75 and that of grade 3 was 0.
66
.

The figure is based on the ROC analysis results of ≥ level 2 (A) and level 3 (B) probabilities obtained by the model
.
Regions of specificity for the largest (>90%) are indicated in blue and the regions for the maximum (>90%) sensitivity are indicated in green

This study showed that radiomic characteristics varied at different grades of PDAC, which increased the accuracy of
CT in preoperative diagnosis.
In this study, a diagnostic model including texture features was established, which could be used to predict the grade of PDAC, which provided support
for preoperative, non-invasive and accurate clinical evaluation.

Original source:

Valeriya S Tikhonova,Grigory G Karmazanovsky,Evvgeny V Kondratyev,et al.
Radiomics model-based algorithm for preoperative prediction of pancreatic ductal adenocarcinoma grade.
DOI:10.
1007/s00330-022-09046-1