Page 155 - Read Online
P. 155
Page 222 Grewal et al. Art Int Surg 2023;3:217-32 https://dx.doi.org/10.20517/ais.2023.28
[53]
preoperative diagnosis and patient evolution; five studies focused on the prediction of complications . Two
of these studies developed imaging-based models, utilizing CT texture features and MRI-based features,
respectively, to predict postoperative pancreatic fistula [54,55] . Capretti et al. similarly harnessed 100
preoperative CT scans to develop a model to predict clinically relevant postoperative pancreatic fistula with
an AUC of 08.07, which was also able to predict postoperative overall complications and length of stays with
[56]
AUCs of 0.690 and 0.709 .
Characterizing tumor biology
One of the most promising applications of radiomics has been its ability to noninvasively assess and
characterize tumor pathology. Considerable work has been done in correlating tumor grade with CT and
MRI radiomic features. The majority of current work has been conducted on radiomics-guided grading in
pancreatic neuroendocrine tumors and has found a remarkable predictive ability for radiomic analysis to
preoperatively distinguish WHO grade 1 from WHO grade 2 or 3 lesions (AUC: 0.736-0.902) [57-61] . Similar
results have been reported for pancreatic cancer (AUC: 0.91-0.994) [62,63] . Radiomic features have been shown
to be able to accurately determine the preoperative risk of lymph node and liver metastases and superior
mesenteric vein and portal vein invasion in pancreatic cancer with an AUC of 0.841-0.912 [51,64-67] . Molecular
analysis can allow for personalized approaches to the management of pancreatic cancer; however, it is
[68]
currently limited by cost and the need for invasive procedures to acquire tissue . Radiogenomics is an area
of great interest where radiomic features can be used to determine the mutational profile of these lesions.
Current studies on radiogenomics in pancreatic cancer have identified key radiomic features that are highly
predictive of genetic alterations in SMAD4, KRAS and p53, HNF1a, and KRT81 status [69-72] .
Predicting treatment response and prognostication
As discussed earlier, real-time assessment of treatment response is essential in the management of
pancreatic cancer. However, currently available clinical tools fail to provide accurate assessment, resulting in
delays in tailoring of therapy that enables us to derive maximum benefit from the administered therapies
while minimizing toxicity. In this regard, studies have assessed radiomic features in a longitudinal manner
[73]
across chemotherapy, known as delta radiomics . These studies have reported a strong association between
delta radiomics and treatment response gauged by changes in CA19-9, RECIST criteria, and grade of
response observed on histopathological examination of the surgical specimen [74-79] . However, current studies
are limited by the heterogeneity of the radiomic features found to be significantly associated with treatment
[73]
response, the small sample size of the study population, and the single institution nature of these studies .
Future studies must consolidate current evidence by analyzing the various radiomic signatures used across
the studies to create a core set of validated radiomics signatures that can be broadly translated to clinical
practice. Lastly, radiomics has shown promise in prognostication. Radiomic models based on features
extracted from pretreatment scans have been shown to be highly predictive of overall survival, local control,
and recurrence and have outperformed current clinical prediction models with an AUC of 0.78-0.92 [80-83] .
APPLICATION OF RADIOMICS IN HEPATOBILIARY MALIGNANCIES
Diagnosis and surgical resectability
Unlike most solid cancers, the diagnosis of HCC & intrahepatic cholangiocarcinoma (ICC) can be
[84]
established without biopsy confirmation . As a result, the role of noninvasive imaging is of particular
importance in the diagnostic workflow of these malignancies. While HCC most often presents with highly
characteristic and specific imaging findings, it can occasionally be challenging to discern definitively on
imaging from other mimicking hypervascular lesions such as hemangiomas, adenomas, and focal nodular
hyperplasia (FNH) [85,86] . Radiomic-guided studies have demonstrated efficacy in differentiating benign liver
lesions from malignant ones with an AUC of differentiating FNH from HCC ranging from 0.917 to 0.971,
an AUC for differentiating hemangiomas from HCC of 0.83-0.96 and a sensitivity and specificity for