Radiomics in pancreatic cancer for oncologist: Present and future

Carolina de la Pinta

Radiation Oncology Department, Ramón y Cajal University Hospital, IRYCIS, Alcalá University, 28034 Madrid, Spain

Keywords:Pancreatic cancer Radiomics Radiogenomics

ABSTRACT Radiomics is changing the world of medicine and more specifically the world of oncology.Early diagnosis and treatment improve the prognosis of patients with cancer.After treatment, the evaluation of the response will determine future treatments.In oncology, every change in treatment means a loss of therapeutic options and this is key in pancreatic cancer.Radiomics has been developed in oncology in the early diagnosis and differential diagnosis of benign and malignant lesions, in the evaluation of response,in the prediction of possible side effects, marking the risk of recurrence, survival and prognosis of the disease.Some studies have validated its use to differentiate normal tissues from tumor tissues with high sensitivity and specificity, and to differentiate cystic lesions and pancreatic neuroendocrine tumor grades with texture parameters.In addition, these parameters have been related to survival in patients with pancreatic cancer and to response to radiotherapy and chemotherapy.This reviewaimed to establish the current status of the use of radiomics in pancreatic cancer and future perspectives.

Introduction

Pancreatic adenocarcinoma is the fourth leading cause of cancer death with a 5-year survival of less than 20%.This is due to late clinical presentation.When diagnosed, the tumor is commonly in advanced or metastatic stage [1].

Although its etiology is not well understood, there are numerous risk factors involved such as a family history of pancreatic cancer, obesity, chronic pancreatitis, exposure to industrial compounds, smoking, preneoplastic lesions or hereditary syndromes of high penetrance associated with pancreatic cancer [1].

Symptoms of the disease depend on the stage and location of the tumor.Pancreatic head tumors account for 60%-75% of cases,and body or tail tumors for 20%-25%.The pathological diagnosis is sometimes based on cytology and the radiological diagnosis is complex.Radiological studies with vascular reconstruction can be useful to establish the possibility of resectability.

Treatment of pancreatic adenocarcinoma should be assessed by a multidisciplinary team with extensive experience.In the case of localized pancreatic lesions, surgery is the only chance of cure.However, at diagnosis only 15%-20% of patients have resectable tumors, 30%-40% are unresectable and 40% have distant metastases [2].

In order to reduce the risk of recurrence after surgery, different neoadjuvant and adjuvant treatment schemes with chemotherapy or radio-chemotherapy have been proposed, despite that 5-year overall survival rates remain low, around 25%-30% in node-negative patients and 10% in nodal disease.In recent years, other local therapeutic modalities have emerged, such as stereotactic body radiation therapy (SBRT), a high-precision irradiation technique that allows very high doses to be administered to the tumor in a limited number of fractions, with a highly cytotoxic ablative biological effect, in addition to microvascular and stromal damage in tumor tissues, boosting the anti-tumor immune response and producing destruction of metastases far from the irradiated lesions [ 3 , 4 ].The advantage over conventional radiotherapy is the administration of high radiation doses with excellent tolerance of surrounding healthy tissues, with adequate local control without limiting the delay or interruption of systemic treatment, improving pain,and preserving the quality of life of patients [5].SBRT in pancreatic tumors is applicable in inoperable tumors, unresectable recurrences, borderline or locally advanced tumors as definitive treatment or with neoadjuvant intent.Its use in adjuvant treatment has also been postulated, although there is less evidence [6].The evaluation of radiological response after these treatments is extremely complex because of the inflammation and necrosis caused by radiotherapy.

Radiomics

The development of radiomics in oncology has played a very important role in the analysis of medical images including computed tomography (CT), magnetic resonance imaging (MRI) andpositron tomography (PET) using computer software.This will facilitate a more precise delimitation of tumor, the tumor microenvironment or alterations after treatment.Radiomics makes it possible to analyze and extract data from medical images, including quantitative and qualitative characteristics.This requires image acquisition, dataset creation, export of DICOM studies, identification of the volume of interest using segmentation tools, feature extraction and qualification, study of the data, construction of a predictive model, and validation of the created models [7].In tumor delimitation several segmentation algorithms are available, including multi-organ atlas-based, landmark-based, shape model-based, and neural network-based [8].

The use of convolutional neural networks (CNNs) in medical image analysis is growing, outperforming traditional machine learning (ML) algorithms on large datasets.However, the large variability in medical concepts poses a limitation to their use.As an alternative, pre-training is proposed with promising results for different image analysis tasks [9].However, few studies have investigated the usefulness of deep features in medical imaging [10–12].Radiogenomics is the correlation of studies between genomics and molecular measurements and radiological studies to be established, improving diagnosis and patient stratification [13].

Radiomics in pancreatic cancer

Early diagnosis and differential diagnosis in pancreatic cancer are essential, as it is a tumor with high mortality and limited therapeutic options, especially in advanced stages.Furthermore, the evaluation of response to treatment is complex because of the inflammation and necrosis caused by radiotherapy using conventional radiological methods [14].

Differential diagnosis in the pancreatic gland

Radiomics is very useful in the differential diagnosis of benign,premalignant and malignant lesions.Radiomics could reduce the need for invasive procedures.

To differentiate autoimmune pancreatitis from pancreatic cancer, MRI is useful.However, it is sometimes not sufficient and may overlap because both entities produce focal mass, pancreatic atrophy, ductal dilatation and common bile duct stenosis [15].Cheng et al.[16]analyzed PET images of pancreatic tumors including morphology, and first order features differentiating autoimmune pancreatitis and pancreatic cancer with a sensitivity of 90.6% and a specificity of 84%.Park et al.[17]analyzed CT images to differentiate autoimmune pancreatitis from pancreatic adenocarcinoma, and showed that radiomics based on texture analysis could help differentiate these two entities with a sensitivity of 89.7%, a specificity of 100% and an overall accuracy of 95.2%.

Chu et al.[18]used radiomic features of CT images to differentiate pancreatic adenocarcinoma and normal pancreatic tissues in a series of patients with a radiological and pathological diagnosis, and the study included a training cohort and a validation cohort.Accuracy, sensitivity and specificity were calculated.Patients were classified with a sensitivity of 100% and a specificity of 98.5%.This would allowa more precise definition of tumor areas, which is very important to local treatment strategies.

These studies support the usefulness of radiomics for the differential diagnosis of pancreatic diseases.

Stratification in cystic lesions

The prevalence of pancreatic cystic lesions is 2.4% to 13.5% [19].The classification of these lesions is very important and some of them may be premalignant lesions.Prevention and early detection programs exist.Radiomics could facilitate the classification of these lesions by establishing an adequate follow-up and treatment.

Cystic lesions can be classified into two categories, neoplastic and non-neoplastic cystic lesions.Pseudocysts are the most frequent non-neoplastic cysts associated with edematous interstitial pancreatitis [20].Cystic neoplasms include serous cystic neoplasm(SCN) (serous cystadenoma), mucinous cystic neoplasm (MCN)(mucinous cystadenoma, cystic mucinous neoplasm with moderate dysplasia and mucinous cystadenocarcinoma), intraductal papillary mucinous neoplasm (IPMN) (intraductal papillary mucinous adenoma and intraductal papillary mucinous carcinoma), and solid pseudopapillary neoplasm (SPN) (pseudopapillary neoplasm and solid pseudopapillary carcinoma) [21].

Radiological identification of the type of cyst varies from 60%to 70% among expert radiologists [22].Radiological variables have been studied to classify these lesions.

Dmitriev et al.[23]differentiated four types of cysts by combining demographic variables with radiomic characteristics of intensity and shape, achieving differentiation of 84% of the lesions.Wei et al.[24]analyzed cyst images in preoperative tests to differentiate SCNs from other pancreatic cystic lesions (PCLs) including 17 intensity and texture features (T-range, wavelet intensity,T-median, and wavelet neighbourhood gray-tone difference matrix busyness) and clinical features.Adequate classification was achieved in 76% of patients and 84% in a validation cohort of 60 patients.Yang et al.[25]evaluated variable slice images, 2 and 5 mm, without affecting feature extraction.In the validation group the accuracy was 74% in patients with 2-mm slice and 83% in 5-mm slice.Hanania et al.[26]stratified patients with IPMN into two groups on preoperative CT scans based on radiomic features,cysts with low-grade dysplasia and cysts with high-grade dysplasia or malignancy, achieving a discrimination between groups of 96%.Permuth et al.[27]analyzed 38 patients with IPMN using 112 radiomic features and 800 microRNAs.The combination of the two features allowed a 92% accuracy in group differentiation.Chakraborty et al.[28]analyzed preoperative CT images to differentiate low-risk and high-risk IPMN in 103 patients ( Table 1 ).

Table 1Radiomics studies in differentiation of pancreatic cystic lesions.

Few published studies have demonstrated the usefulness of radiomics in differentiating potentially malignant from benign cystic lesions [20–27].

Pancreatic neuroendocrine tumors (PNETs)

PNETs are a heterogeneous group of tumors with varying degrees of aggressiveness and account for 12.1% of all digestive neuroendocrine tumors [29].

Lin et al.[30]analyzed clinical and textural features, entropy,skewness, kurtosis and uniformity to differentiate PNETs and intrapancreatic accessory spleen.There were no differences in margin, degree of enhancement, lymph nodes or size between PNET and intrapancreatic accessory spleen (allP>0.05).Patients with intrapancreatic accessory spleen showed heterogeneous enhancement in arterial phase and the same degree of enhancement as the spleen in portal phase, both greater than those in PNET (P= 0.06;P= 0.04).Entropy and uniformity were different (P<0.01).Analysis showed that uniformity at moderate and high sigma had high area under the curve (0.82 and 0.89) with better sensitivity (85%-95%) and acceptable specificity (75%-83.3%) in intrapancreatic spleen differentiation.Concluding that entropy and uniformity were significantly different between intrapancreatic accessory spleen and pancreatic neuroendocrine tumor with moderate and high sigma value withPvalue<0.01 (high sensitivity of 85%-95%and specificity of 75%-83.3%) [30].

The differentiation of PNET grades was studied by Choi et al.[31].Predictors of worse grade (grade 2/3) were a well-defined margin [odds ratio (OR) = 7.273], low sphericity(OR = 0.409) in 2D arterial analysis, high skewness (OR = 1.972)and low sphericity (OR = 0.408) in 3D analysis, low kurtosis(OR = 0.436) and low sphericity in 2D portal (OR = 0.420), and large surface area (OR = 2.007) and low sphericity in 3D portal analysis (OR = 0.503) (P<0.05) [31].

Canellas et al.[32]also analyzed clinical and radiomic features by CT texture to predict PNET grade.The authors concluded that texture analysis and CT features are predictive of PNET aggressiveness and can be used to identify patients at risk of early progression prior to surgical resection.

Li et al.[33]studied CT texture to differentiate atypical PNETs from pancreatic adenocarcinoma.Texture parameters including mean, median, 5th, 10th, 25th, 75th and 90th percentiles, skewness, kurtosis and entropy were analyzed.Sixty-seven percent of PNETs were typical hypervascular and 32% atypical hypovascular.The authors concluded that for differentiating pancreatic adenocarcinoma and atypical PNETs, the 5th percentile and 5th percentile plus skewness were optimal parameters alone or in combination for the diagnosis of pancreatic adenocarcinoma and atypical PNETs respectively and could help to distinguish atypical PNETs from adenocarcinomas [33].Other authors had studied differentiation between the different grades [ 34 , 35 ]( Table 2 ).

Table 2Radiomics studies in PNETs.

Pancreatic cancer

The use of radiomics has been studied in predicting survival and local control in pancreatic cancer patients.

Yue et al.calculated and identified texture variations in PET before and after radiotherapy in 26 pancreatic cancer patients by classifying patients into low- and high-risk with long and short overall survival (OS), respectively, using texture parameters such as standardized uptake value max, homogeneity, variance, sum mean and cluster tendency [36].Cassinotto et al.analyzed attenuation and Yun et al.extracted second-order radiomic features such as histograms and gray level co-occurrence matrix (GLCM) from preoperative CT images [ 37 , 38 ].Cozzi et al.[39]identified a pre-SBRT CT-based radiomic signature based on GLCM, gray level run length matrix (GLRLM), neighborhood gray level different matrix(NGLDM) and gray level zone length matrix (GLZLM) that correlated with OS and local control (LC) after SBRT.In low-risk patients, the median OS and LC in the validation group were 14.4 and 28.6 months; while in high-risk patients, they were 9 and 17.5 months, respectively.The concordance index was 0.69 ±0.08 for training and 0.75 ±0.10 for validation.Eilaghi et al.[40]investigated CT textures with respect to OS in pancreatic adenocarcinoma in 30 patients finding a signature of 5 features associated with OS.

Sandrasegaran et al.[41]analyzed texture in prognostic prediction in pancreatic cancer in 70 patients with unresectable pancreatic cancer.Mean, kurtosis, entropy, skewness in analysis, arterial and venous invasion, metastatic disease and tumor size were included and correlated with OS and progression-free survival (PFS).Median survival and PFS were 13.3 and 7.8 months, respectively.Authors concluded that OS was associated with texture features in pancreatic cancer particularly for patients undergoing curative intent surgical resection.

Chen et al.[42]analyzed radio-induced changes in quantitative CT features of 20 cancer patients after radiochemotherapy (CRT).They observed changes in these features that can be used to assess early response to treatment and intensify treatment in nonresponders or poor responders.

These findings are very important for defining therapeutic strategies in pancreatic cancer patients prior to treatment.If there is certainty of local progression, the oncologist would start or change a treatment, but this is difficult to assess today with the available imaging.

Assessment of response to treatment

Changes after treatment with chemotherapy and/or radiotherapy are difficult to determine by conventional methods and therefore, radiomic features are being studied.

After radiotherapy treatment, some parameters such as histogram shape, energy and entropy have been detected.However, the mechanisms of change are not clear.Patients with good response to treatment tend to have large reductions in mean CT number (MCTN) and skewness and increases in standard deviation and kurtosis.These metrics could help identify responders from non-responders by tailoring treatments according to response.Chen et al.[42]evaluated the usefulness of radiomics in response assessment during CRT treatment in pancreatic cancer.Eight parameters including MCTN, peak position,volume, standard deviation, skewness, kurtosis, energy and entropy were analyzed in each fraction of 20 patients.The decrease in volume, increase in skewness, decrease in kurtosis,and reduction in MCTN became significant after two weeks of treatment.Pathological response was associated with changes in MCTN, standard deviation, and skewness.In cases of treatment response, patients tended to have large reductions in MCTN and skewness, and large increases in standard deviation and kurtosis.

Ciaravino et al.[43]analyzed CT textures in patients with pancreatic adenocarcinoma treated with neoadjuvant chemotherapy followed by surgery as a post-treatment evaluation.Seventeen patients with unresectable and borderline pancreatic adenocarcinoma were included.Comparison between pre- and post-treatment kurtosis showed significant differences, and the authors concluded that radiomics could be used as a post-neoadjuvant assessment tool in patients without evidence of response by conventional methods.

However, some studies need validation of their results [ 42 , 43 ].This may be because these clinical outcomes are influenced by several factors, including pre- and post-CRT effects and the size included too small and the population too heterogeneous.We need more evidence.

Other utilities of radiomics in pancreatic cancer treatment

The delineation of treatment volumes in radiotherapy is key to correct treatment, and this is more important by modern radiotherapy and techniques such as SBRT.Bian et al.analyzed the relationship between radiomics and nodal involvement in pancreatic cancer [44].They evaluated 225 patients with resected pancreatic cancer with CT within 1 month after resection.They analyzed 1029 radiomic features on arterial phase CT and were able to define the presence and absence of nodal involvement.Ji et al.used radiomics to predict the presence of nodal involvement [45].They showed good model calibration and discrimination in the training and validation cohorts, which could become a contouring tool in radiotherapy.

Radiogenomics in pancreatic cancer

The genetic study of tumors has allowed the development of targeted therapies, and the stratification of patients in terms of risk of relapse, prognosis, prediction of response and survival.This breakthrough will allow the use of increasingly less invasive tools.From this effort comes radiogenomics, the correlation of genetic alterations or tumor microenvironment and radiological findings allows the use of these imaging tests as a non-invasive tool for personalized medicine.

Attiyeh et al.[46]analyzed the relationship between radiomic variables and tumor genotype in pancreatic cancer.The follow-ing genes were included:KRAS,TP53,CDKN2A, andSMAD4.In 35 patients, 34 hadKRAS, 29 hadCDKN2A, 16 hadSMAD4and 29 hadTP53expressions.The number of altered genes had predictive value of OS (P= 0.016).Strong differentiation was found between tumors with and withoutSMAD4alterations.There seems to be a relationship between higher number of mutations and heterogeneity in the image.The authors demonstrated that radiomic features extracted from CT are associated with genotype, number of altered genes and stromal content in pancreatic cancer.These associations could help to develop survival prediction tools.

Technical problems with radiomics, limitations and challenges

Despite all the available evidence, radiomics has limitations.Some of the factors that may affect radiomic results include: CT acquisition, reconstruction, kernels, tube currents, slice size, voxel size, grey level, presence and delay of contrast enhancement [47].

Yamashita et al.demonstrated that differences in contrastenhanced CT acquisition affected the results of the radiomic study leading to changes in segmentation and its reproducibility and comparability between series [48].The study did not demonstrate statistically significant differences in CT model, pixel spacing, and contrast administration ratio.The study suggests that radiologists are more or less sensitive to CT acquisition parameters, demonstrating the importance of adjusting for these variables to established protocols.Furthermore, this study support the hypothesis of the usefulness of a semi-automated segmentation tool previously trained by several radiologists that can homogenize these variations.Standardization of protocols is therefore important, in addition to external validation [ 49 , 50 ].Also many of the comparisons between diagnostic entities using radiomics are subjective and not clinically applicable.For example, the distinction between pancreatic adenocarcinoma and pancreatic neuroendocrine tumors alone.

Conclusion

Radiomics is a promising non-invasive tool for the diagnosis and clinical management of pancreatic tumors.The usefulness of radiomics has been studied in the differential diagnosis of benign,premalignant and malignant lesions in the pancreas.In addition,in patients with neoadjuvant pancreatic cancer, it can help in the more precise definition of lesions for radiotherapy and assessment of response.Radiomics provides a more adequate and reproducible measurement of the tumor than other methods.In addition, the combination of radiomics and genomics has a promising future.However, image acquisition protocols and radiomic analysis systems need to be standardized and validation cohorts are needed.Further studies are needed to consolidate the available data.

Acknowledgments

We thank the members of the IRYCIS for their help and advice throughout this research.

CRediT authorship contribution statement

Carolina de la Pinta:Conceptualization, Data curation, Formal analysis, Methodology, Supervision, Writing –original draft, Writing –review & editing.

Funding

None.

Ethical approval

Not needed.

Competing interest

No benefits in any form have been received or will be received from a commercial party related directly or indirectly to the subject of this article.