An artificial intelligence deep learning model for identification of small bowel obstruction on plain abdominal radiographs.

OBJECTIVES: Small bowel obstruction is a common surgical emergency which can lead to bowel necrosis, perforation and death. Plain abdominal X-rays are frequently used as a first-line test but the availability of immediate expert radiological review is variable. The aim was to investigate the feasibility of using a deep learning model for automated identification of small bowel obstruction. METHODS: A total of 990 plain abdominal radiographs were collected, 445 with normal findings and 445 demonstrating small bowel obstruction. The images were labelled using the radiology reports, subsequent CT scans, surgical operation notes and enhanced radiological review. The data were used to develop a predictive model comprising an ensemble of five convolutional neural networks trained using transfer learning. RESULTS: The performance of the model was excellent with an area under the receiver operator curve (AUC) of 0.961, corresponding to sensitivity and specificity of 91 and 93% respectively. CONCLUSION: Deep learning can be used to identify small bowel obstruction on plain radiographs with a high degree of accuracy. A system such as this could be used to alert clinicians to the presence of urgent findings with the potential for expedited clinical review and improved patient outcomes. ADVANCES IN KNOWLEDGE: This paper describes a novel labelling method using composite clinical follow-up and demonstrates that ensemble models can be used effectively in medical imaging tasks. It also provides evidence that deep learning methods can be used to identify small bowel obstruction with high accuracy.

Routes to diagnosis and missed opportunities in the detection of renal cancer.

AIM: To determine the proportion of renal cancers for which incidental detection was the route to diagnosis, the characteristics of tumours identified in this way, and the frequency with which opportunities to make this diagnosis were missed. MATERIALS AND METHODS: Consecutive patients with renal cancers treated at Royal Cornwall Hospitals NHS Trust (April 2011 and July 2018 inclusive) were identified from the Trust's cancer registry database, and a retrospective review of the imaging and electronic case notes was undertaken. Mann-Whitney U-tests for comparison of patient age and tumour size at diagnosis, and chi-squared tests for comparing cell type distribution and grade were performed. Logistic regression was then used to identify the characteristics of patients in whom a renal tumour was missed initially. RESULTS: Of 327 patients, 194 (63%) presented incidentally, and 133 (37%) symptomatically. Incidentally detected cancers were found in younger patients, (median of 65 years versus 69 p=0.01) and were smaller at presentation (median of 5.5 versus 7.2 cm, p<0.00001). Thirty-six different reporters missed opportunities to diagnose renal cancer in 50 (16%) patients on 78 occasions, 28 lesions (35%) being missed more than once. Thirty were imaged incompletely; four were visible only on a single image and three on a scout view at magnetic resonance imaging. CONCLUSION: The commonest route to diagnosis of renal cancer is by incidental detection of a mass. In 16% of patients in whom renal cancer is diagnosed, there is at least one prior examination on which the lesion is visible at an earlier date. The clinical impact of these missed diagnostic opportunities remains uncertain.

Getting it right first time in radiology.

Getting it Right First Time (GIRFT) is a national programme designed to improve medical care in the National Health Service (NHS) in England by reducing unwarranted variation. By tackling variation in the way services are delivered across the NHS and by sharing best practice, GIRFT identifies changes that will help improve care and patient outcomes as well as delivering efficiencies.

The accuracy of interpretation of emergency abdominal CT in adult patients who present with non-traumatic abdominal pain: results of a UK national audit.

AIM: To evaluate major/minor discrepancy rates for provisional (initial) and addendum (supplementary senior review) emergency computed tomography (CT) reports in patients presenting with non-traumatic abdominal pain. MATERIALS AND METHODS: Ethical approval for this type of study is not required in the UK. All radiology departments with an approved lead for audit registered with the Royal College of Radiologists were invited to participate in this retrospective audit. The first 50 consecutive patients (25 surgical, 25 non-surgical) who underwent emergency abdominal CT for non-traumatic abdominal pain in 2013 were included. Statistical analyses were performed to identify organisational and report/patient-related variables that might be associated with major discrepancy. RESULTS: One hundred and nine (58%) of 188 departments supplied data to the study with a total of 4,931 patients (2,568 surgical, 2,363 non-surgical). The audit standard for provisional report major discrepancy was achieved for registrars (target <10%, achieved 4.6%), for on-site consultants (target <5%, achieved 3.1%) and consultant addendum (target <5%, achieved 2.9%). Off-site reporters failed to meet the standard target (<5%, achieved 8.7% overall and 12.7% in surgical patients). The standard for patients coming to harm was not met in the surgical group (target <1%, achieved 1.5%) and was narrowly missed overall (target <1%, achieved 1%). CONCLUSION: This study should be used to provide impetus to improve aspects of out-of-hours CT reporting. Clear benefits of CT interpretation/review by on-site and more senior (consultant) radiologists have been demonstrated.

Multicentre survey of radiologically inserted gastrostomy feeding tube (RIG) in the UK.

AIMS: To evaluate the variance in current UK clinical practice and clinical outcomes for direct percutaneous radiologically inserted gastrostomy (RIG). MATERIALS AND METHODS: A prospective UK multicentre survey of RIG performed between October 2008 and August 2010 was performed through the British Society of Gastrointestinal and Abdominal Radiology (BSGAR). RESULTS: Data from 684 patients were provided by 45 radiologists working at 17 UK centres. Two hundred and sixty-three cases (40%) were performed with loop-retained catheters, and 346 (53%) with balloon-retained devices. Sixty percent of all patients experienced pain in the first 24 h, but settled in the majority thereafter. Early complications, defined as occurring in the first 24 h, included minor bleeding (1%), wound infection (3%), peritonism (2%), and tube misplacement (1%). Late complications, defined as occurring between day 2 and day 30 post-procedure, included mild pain (30%), persisting peritonism (2%), and 30 day mortality of 1% (5/665). Pre-procedural antibiotics or anti-methicillin-resistant Staphylococcus aureus (MRSA) prophylaxis did not affect the rate of wound infection, peritonitis, post-procedural pain, or mortality. Ninety-three percent of cases were performed using gastropexy. Gastropexy decreased post-procedural pain (p < 0.001), but gastropexy-related complications occurred in 5% of patients. However, post-procedure pain increased with the number of gastropexy sutures used (p < 0.001). The use of gastropexy did not affect the overall complication rate or mortality. Post-procedure pain increased significantly as tube size increased (p < 0.001). The use of balloon-retention feeding tubes was associated with more pain than the deployment of loop-retention devices (p < 0.001). CONCLUSION: RIG is a relatively safe procedure with a mortality of 1%, with or without gastropexy. Pain is the commonest complication. The use of gastropexy, fixation dressing or skin sutures, smaller tube sizes, and loop-retention catheters significantly reduced the incidence of pain. There was a gastropexy-related complication rate in 5% of patients. Neither pre-procedural antibiotics nor anti-MRSA prophylaxis affected the rate of wound infection.

Case report: Thromboembolic and septic complications of migrated cyanoacrylate injected for bleeding gastric varices.

Endoscopic injection of N-butyl-2-cyanoacrylate combined with radio-opaque lipiodol is widely used to achieve haemostasis in bleeding gastric varices. We present a case of migration of injected cyanoacrylate, thrombus formation and subsequent septic embolisation.

CT colonography: interpretative performance in a non-academic environment.

AIM: To investigate interpretative accuracy and reporting time for radiologists performing computed tomography (CT) colonography in day-to-day non-academic clinical practice. MATERIALS AND METHODS: Thirteen radiologists from seven centres, who were reporting CT colonography in non-academic daily clinical practice, interpreted a dataset of 15 colonoscopically validated cases in a controlled environment. Ten cases had either a cancer or polyp >10mm; one case had a medium polyp and four were normal. Correct case categorization and interpretation times were compared using analysis of variance to aggregated results obtained from both experienced observers and observers recently trained using 50 cases, working in an academic environment. The effect of experience was determined using Spearman's rank correlation. RESULTS: Individual accuracy was highly variable, range 53% (95% CI 27-79%) to 93% (95% CI 68-100%). Mean accuracy overall was significantly inferior to experienced radiologists (mean 75 versus 88%, p=0.04) but not significantly different from recently trained radiologists (p=0.48). Interpretation time was not significantly different to experienced readers (mean 12.4 min versus 11.7, p=0.74), but shorter than recently trained radiologists (p=0.05). There was a significant, positive, linear correlation between prior experience and accuracy (p<0.001) with no plateau. CONCLUSION: Accuracy for sub-specialist radiologists working in a non-academic environment is, on average, equivalent to radiologists trained using 50 cases. However, there is wide variability in individual performance, which generally falls short of the average performance suggested by meta-analysis of published data. Experience improves accuracy, but alone is insufficient to determine competence.

An unusual complication of oesophageal stent deployment.

Insertion of a self-expanding metal stent is commonly used to palliate dysphagia secondary to oesophageal carcinoma. Displacement of the stent itself is a recognized complication which can result in perforation of the upper gastrointestinal tract. We report the first case of bowel obstruction and perforation resulting from the olive of the deployment system. This was probably due to peritoneal deposits on the small intestine obstructing passage of the olive.

Pineapple juice as a negative oral contrast agent in magnetic resonance cholangiopancreatography: a preliminary evaluation.

The quality of magnetic resonance cholangiopancreatography (MRCP) images is frequently degraded by high signal from the gastrointestinal tract. The aim of this study is to evaluate pineapple juice (PJ) as an oral negative contrast agent in MRCP. Preliminary in vitro evaluation demonstrated that PJ shortened T(2) relaxation time and hence decreased T(2) signal intensity on a standard MRCP sequence to a similar degree to a commercially available negative contrast agent (ferumoxsil). Electrothermal atomic absorption spectrometry assay demonstrated a high manganese concentration in PJ of 2.76 mg dl(-1), which is likely to be responsible for its T(2) imaging properties. MRCP was subsequently performed in 10 healthy volunteers, before and at 15 min and 30 min following ingestion of 400 ml of PJ. Images were assessed blindly by two Consultant Radiologists using a standard grading technique based on contrast effect (degree of suppression of bowel signal), and image effect (diagnostic quality). There were statistically significant improvements in contrast and image effect between pre and post PJ images. There was particularly significant improvement in visualization of the pancreatic duct, but no significant difference between 15 min and 30 min post PJ images. Visualization of the ampulla, common bile duct, common hepatic and central intrahepatic ducts were also significantly improved at 15 min following PJ. Our results demonstrate that PJ, may be used as an alternative to commercially available negative oral contrast agent in MRCP.