RPS 701 - Upper GI tract: what is new?

Purpose:

This study used CT perfusion imaging (CTPI) technology to evaluate the diagnostic efficacy in differentiating metastatic from inflammatory perigastric lymph nodes in patients with gastric cancer.

Methods and materials:

A total of 115 perigastric lymph nodes of 60 patients with gastric cancer confirmed by gastroscopy underwent CTPI scan before an operation. The scan data was post-processed by using commercial software to calculate perfusion parameters including blood flow (BF) and permeability surface (PS), and to measure the size of lymph nodes. According to the postoperative pathology result, the lymph nodes were divided into two groups: metastatic nodes and inflammatory nodes. Perfusion parameters and the size of lymph nodes between two groups were respectively compared statistically, and a ROC analysis was used to determine the optimal cutoff value with sensitivity, specificity, and area under the curve (AUC).

Results:

Examined perigastric lymph nodes were metastatic in 65 and inflammatory in 50. Mean values of perfusion parameters in metastatic and inflammatory lymph nodes, respectively, were BF of 90.05 versus 79.31 ml/100 mg /min (p < 0.01), and PS of 42.19 vs. 35.89 ml/100 mg /min (p < 0.01). Mean values of the size in metastatic and inflammatory lymph nodes were 1.34 cm versus 1.16 cm (p > 0.05). The sensitivity of 81.5%, the specificity of 66.0 %, and AUC of 0.784 for BF with a cutoff value of 80.76 ml/100 mg /min for differentiating metastatic from inflammatory nodes were higher than those of PS or the size of lymph nodes (P <0.05).

Conclusion:

BF might be a more effective marker than PS or the size of lymph nodes for differentiating metastatic from inflammatory nodes in gastric cancer patients.

Limitations:

The unknown cardiac output of patients might influence CT perfusion values.

Ethics committee approval

n/a

Funding:

No funding was received for this work.

Purpose:

To establish the role of our gastric CT protocol, with additional virtual gastroscopy navigation, in the staging of gastric cancer.

Methods and materials:

Between August 2018 and September 2019, we selected 34 patients (mean age 64.2, 58.82% male) and scanned them in our 160 detector row system, using a dedicated CT protocol with standard gastric fluid distention obtained with a 2 litre diluted sodium amidotrizoate and meglumine amidotrizoate solution. The studies were reviewed by two radiologists with gastrointestinal and abdominal imaging experience, either in standard 2D or 3D navigation mode. The results were then compared for accuracy, specificity, and sensitivity with the pathology data.

Results:

The lesions we detected were divided in three T groups: T1/T2 (44.11%), T3 (26.47%), and T4 (29.42%). Sensitivity values were 77.9% for T1/T2, 87.1% for T3, and 100% for T4 lesions. Specificity values were 100% for T1/T2, 83.2% for T3, and 92.3% for T4 lesions. Global accuracy values were 85.9% for T1/T2, 82.2% for T3, and 91.0% for T4 lesions. Nodal status was evaluated as well, with 23 N-positive patients (73.52%); nodal staging sensitivity, specificity, and accuracy values were 91.0%. 64.2%, and 75.7%.

Conclusion:

Our CT protocol with fluid distention of the stomach and 3D virtual navigation has shown to have high sensitivity, specificity, and overall global accuracy in the staging of gastric cancer, especially for the T status in the early stages (T1/T2) and nodal status, allowing a more accurate therapeutic planning, especially concerning patients eligible for surgery.

Limitations:

Our study is monocentric and based on a relatively small number of patients (34).

Ethics committee approval

All patients were orally instructed and signed written informed consent obtained.

Funding:

No funding was received for this work.

Purpose:

To evaluate the diagnostic accuracy of contrast-enhanced MRI in the differentiation of mucosal high-grade neoplasia (MHN) from early invasive squamous cell cancer (EISCC) of the oesophagus.

Methods and materials:

32 study participants with MHN (n=18) and EISCC (n=14) of the oesophagus were enrolled in this prospective study. Postoperative histopathologic analysis was the reference standard. All participants underwent MRI (MAGNETOM Skyra, Siemens Healthineers, Erlangen, Germany). Two radiologists, blinded to participants’ data, independently evaluated enhancement degree and enhancement pattern. Diagnostic performance of the two features was compared using the Chi-square test; kappa values were assessed for reader performance.

Results:

Surgery was performed within 3.8+2.5 days after MR imaging. Inter-reader agreement on image quality was excellent (Kappa value=0.801~0.902 for three sequences, P<0.001), and the agreement on MR features was also excellent (Kappa value=0.831, P<0.001). All 4 mass-like MHN were “heart-shaped” in appearance. The combined two features provided the best sensitivity, specificity, and AUC of 100%, 94.1%, and 0.987, respectively. There was no significant difference between the degree of enhancement and the combination of the two features.

Conclusion:

Contrast MRI can differentiate MHN from EISCC in oesophagus, with the presence of “heart-shaped” appearance favouring the diagnosis of MHN.

Limitations:

Limited sample size, because most MHN was removed by EMR/ESD. Dynamic enhanced images features were not analysed in this study because it remains difficult to acquire good image quality of dynamic enhanced images of the chest.

Ethics committee approval

This study was approved by the Hospital Ethics Committee and all participants signed the written informed consent form.

Funding:

No funding was received for this work.

Purpose:

To determine the accuracy of staging duodenal adenocarcinoma (DA) with contrast-enhanced CT. DA represents over 50% of small bowel adenocarcinomas. Patients usually present with aspecific symptoms, causing significant delays in diagnosis. Currently, it is unclear what would be the optimal diagnostic workup in DA.

Methods and materials:

Retrospective evaluation of TNM-stage of DA on portal-venous CECT in 50 patients.

Results:

The tumour was visible in 90% (45/50) of patients, isodens in 46% (23/50), hypodens in 30% (15/50), hyperdens in 2% (1/50), and of mixed density in 12% (6/50). Tumours were localised in all segments of the duodenum, but mostly in the pars descendens (22/50=44%). 35 patients underwent surgery with curative intent, 28 were resected, and 7 had unexpected metastases (N=4) or local invasion (N=3). The T-stage of the 28 resected patients was correct in 13 (13/28=46%), overestimated in 2 (2/28=7%), underestimated in 10 (10/28=36%), and in 3 patients the T-stage could not be determined. Of the 28 resected patients, 21 had suspicious lymph nodes (LNs), but only 13 of these had positive LNs at pathology. Of the 7 patients without suspicious LNs, 3 had positive LNs at pathology. In 12% (6/50), tumours in the duodenum were missed on endoscopy. In 27 patients with MRI, tumour visibility and extension could be better determined on the HASTE than on CECT.

Conclusion:

CECT is probably not sufficient for staging DA. Compared to pathology, CECT underestimates the T-stage and cannot reliably detect LN metastases, which are highly prognostic. Future research concerning additional imaging modalities is needed to improve staging DA. We suggest that MRI could be a potential candidate.

Limitations:

Retrospective and therefore non-uniform data.

Ethics committee approval

Informed consent waived and study approved by ethical committee.

Funding:

No funding was received for this work.

Purpose:

A barium swallow is a fluoroscopic test of the upper gastrointestinal tract. It involves the use of ionising radiation and is operator dependent with respect to radiation exposure, which can be reduced with staff training.

In this observational study, 1004 barium swallow cases were evaluated for the relationship between radiation dose and staff level of training, and correlated with the final clinical/histologic diagnoses.

Methods and materials: Results: Conclusion:

This study demonstrates that radiation dose, though largely within the accepted range, is lowest for consultant GI radiologists with no difference in pathology identification. Re-visiting consultant supervised barium training will help further lower the radiation exposure to patients. Barium swallow study remains a valuable imaging modality for evaluating oesophageal pathologies in 21^st century medicine.

Limitations: Ethics committee approval

n/a

Funding:

Purpose:

To establish the role of the dynamic oesophagogram in the grading of achalasia, in comparison with our standard static barium swallow protocol, and evaluate its correlation with clinical/manometric subtypes classification.

Methods and materials:

Between June 2017 and June 2019, we selected 109 patients (mean age 61.4, 56.68% male) and evaluated them, both with our static (SBS) and dynamic barium swallow (DBS) x-ray protocol. The studies were reviewed in blind by two radiologists with gastrointestinal and abdominal imaging experience. The results were then evaluated for accuracy, specificity, and sensitivity, and compared with the manometric/clinical grading, our gold standard.

Results:

The patients were divided into 4 radiological grades for both SBS and DBS, according to morphological criteria (I-IV), corresponding to three clinical subtypes (1-3; subtype 3 accomodating radiological grades III-IV). Sensitivity values ranged between 78.1-100% for SBS and between 96.3-100% for DBS. Specificity values were substantially similar for both SBS and DBS, ranging between 58-97%. Subtype grading accuracy was significantly higher for DBS in Subtype I (82.3 vs 70.7%).

Conclusion:

Our dynamic oesophagogram protocol, compared with our static dynamic swallow, has shown to have similar specificity, better sensitivity, and global accuracy for the diagnosis of achalasia. This has been linked to the much higher sensitivity in dysmotility detection shown by DBS. Grading definition accuracy is particularly high for DBS when compared with the manometric/clinical staging. These results are promising and DBS could help guide therapeutic choices through a more precise patient stratification.

Limitations:

Our study is monocentric and the number of patients is limited (109).

Ethics committee approval

Our study received ethical committee approval and each patient signed informed consent before examinations.

Funding:

No funding was received for this work.

Purpose:

Gastrointestinal stromal tumours (GISTs) are rare mesenchymal tumours of the gastrointestinal (GI) tract and difficult to diagnose. Here, we evaluate radiomics for 1) distinguishing GIST from other GI tumours, and 2) predicting the GIST c-KIT mutational status and mitotic index.

Methods and materials:

Patients treated at the Erasmus MC between 2004-2017 with GIST or non-GIST GI tumours with a venous phase CT scan were included. A clinician segmented the tumours from which 424 radiomics features were extracted. Sex, age, and tumour location were collected. Prediction models for GIST versus non-GIST, presence of c-KIT mutation and type (exon9 or exon11), and mitotic index (≤ 5/50 high power fields (HPF) vs. > 5/50 HPF) were created through an automated search amongst a variety of machine learning algorithms to find the combination that maximises performance. The evaluation was implemented through 100x random-split cross-validation, with 80% of the data used for training and model optimisation and 20% for testing. For comparison, a subset was scored by a radiologist.

Results:

The dataset consisted of 122 non-GISTs (schwannomas, leiomyosarcomas, leiomyomas, gastric cancers, and lymphomas) and 125 GISTs. The GIST versus non-GIST radiomics model had a mean area under the curve (AUC) of 0.72; adding location improved the performance to 0.83; adding age and gender did not improve the performance. The radiologist had a similar performance (AUC of 0.72). The c-KIT mutation models (c-KIT, exon9, and exon11) only predicted the absence of mutations. The mitotic index model had an AUC of 0.60.

Conclusion:

Our radiomics approach proved useful for distinguishing GISTs from other GI tumours, with a performance similar to a radiologist, but could not predict the GIST c-KIT mutational status or mitotic index.

Limitations:

Potential location bias.

Ethics committee approval

Erasmus MC IRB (MEC-2017-1187).

Funding:

NWO (#14929-14930).

Purpose:

To develop a risk model using a contrast-enhancement computed tomography (CECT) based on deep learning method to predict occult peritoneal metastasis (OPM) in patients with gastric cancer (GC).

Methods and materials:

A total of 567 GC patients (training (OPM-positive n=55, OPM-negative n=340) and internal validation (OPM-positive n=24, OPM-negative n=148) cohort) were subject to develop the risk model based on deep learning method (Xception). All patients’ OPM status was diagnosed as negative by CECT and later confirmed by laparoscopy or surgery (OPM-positive n=79, OPM-negative n=488). The region of interest was manually drawn along the margin of the primary tumour and extracted information reflecting phenotypes of the primary tumour. In the training cohort, we used 5-fold cross-validation and fuse for the prediction results obtained separately on the internal validation cohort. Receiver operating characteristics (ROC) and the area under ROC (AUC) were used to measure the model’s discrimination and the decision curve analysis was used to compare the benefit of the model and all-laparoscopy and none-laparoscopy schemes on the internal validation cohort.

Results:

The AUC yield was 0.8273 (median AUC, range in [0.8115-0.9950]), 0.7606 (median AUC, range in [0.7515-0.7727]) and 0.8558 for the cross-validation training cohort, cross-validation validation cohort and internal validation cohort. Decision curve analysis showed that if the threshold probability in the clinical decision was less than 50%, the patient would benefit more from our model than either of the all-laparoscopy or none-laparoscopy schemes.

Conclusion:

Our model has an excellent prediction ability of OPM using only CECT and may have significant clinical implications in the early detection of OPM for GC.

Limitations:

The region of interest was delineated in one single slice.

Ethics committee approval

This retrospective study was approved by the institutional review board of our hospital and informed consent was waived.

Funding:

No funding was received for this work.

Purpose:

To evaluate the association between imaging features of gastrointestinal stromal tumours (GISTs) on preoperative contrast-enhanced computed tomography (CECT) and risk of relapse according to Miettinen's risk classification.

Methods and materials:

The preoperative CECT of all patients with a pathologically proven diagnosis of GIST who underwent surgery between January 2008 and June 2019 were retrospectively reviewed. Exclusion criteria were recurrent GISTs, “paediatric-type” GISTs, and treatment with Imatinib prior to surgery. The following imaging features were analysed: growth pattern (exophytic/endoluminal/mixed), calcifications (presence/absence), necrosis (presence/absence), signs of ulceration or fistulisation (presence/absence), internal foci of haemorrhage (presence/absence), margins (irregular/well-defined), enlarged feeding or draining vessels (presence/absence), type of CE (homogenous/heterogeneous), degree of CE (marked versus mild and moderate), ascites (presence/absence), lymphadenopathy (presence/absence), peritoneal implants, and liver and other organs metastasis (presence/absence). The Chi-Square statistic was used to examine the correlation between different imaging features and pathologic risk classes of surgical resection specimens.

Results:

54 patients (29 men, 25 women; median age 65±11 years; range: 30-84 years) were included in the study for a total of 56 primitive GISTs: 5 (8.9%) no risk, 13 (23.2%) very low risk, 17 (30.4%) low risk, 8 (14.3%) moderate risk, and 13 (23.2%) high risk. Necrosis (p<0.001), ulceration or fistulisation (p=0.004), haemorrhage (p=0.007), margins (p=0.005), enlarged vessels (p<0.001), type of CE (p<0.001), and metastasis (p<0.001) were found to be associated with the Miettinen’s risk group.

Conclusion:

The presence of necrosis, signs of ulceration or fistulisation, internal foci of haemorrhage, irregular margins, enlarged vessels, heterogeneous CE, and metastasis were significantly associated with high risk of relapse and CECT imaging may be useful for risk stratification.

Limitations:

Retrospective and observational study design.

Ethics committee approval

n/a

Funding:

No funding was received for this work.

Purpose:

Neoadjuvant chemoradiotherapy (NCRT) plus surgery improves long-term survival of patients with locally advanced esophageal squamous cell carcinoma (ESCC). Treatment response prediction remains a great challenge. We aimed to evaluate the value of deep learning radiomics models based on computed tomography (CT) for predicting pathologic complete response (pCR) in ESCC patients receiving NCRT.

Methods and materials:

We identified 161 patients with ESCC (mean age: 58, male: 83.5%, pCR: 46.0%). A total of 2048 deep learning radiomics features were analysed by the convolutional neural network (Xception) from CT images. After feature selection, a radiomics signature was built with an extreme gradient boosting (XGBoost) algorithm. Two models were built. Model A, for post-NCRT evaluation, incorporates both pre-NCRT and post-NCRT CT images into the analysis, while Model B, for pretreatment assessment, was built based on pre-NCRT CT images only.

Results:

Model A comprised 9 selected features and showed good discrimination performance in test set for treatment response to NCRT, with an accuracy of 0.78, area under the receiver operating characteristic curve (AUC) of 0.89, sensitivity of 0.70, and specificity of 0.96. Calibration curves demonstrated good agreement between the prediction probability and the observed pCR (Hosmer-Lemeshow test, P-value = 0.66). Decision curve analysis confirmed the clinical benefits. Model B was also found to have predictive potential, with an accuracy of 0.67, AUC of 0.73, sensitivity of 0.87, and specificity of 0.62.

Conclusion:

Deep learning radiomics analysis based on CT demonstrated promising predictive value for NCRT treatment response in locally advanced ESCC. Both pretreatment and post-NCRT models could be potentially used for treatment strategy decision-making.

Limitations:

Future studies in larger prospective cohorts are needed to further confirm clinical practicability.

Ethics committee approval

n/a

Funding:

Supported by Grant No. 179 from the Health Ministry of China.