Search

Artificial intelligence (AI) and digital innovation are transforming healthcare. Technologies such as machine learning in image analysis, natural language processing in medical chatbots and electronic medical record extraction have the potential to improve screening, diagnostics and prognostication, leading to precision medicine and preventive health. However, it is crucial to ensure that AI research is conducted with scientific rigour to facilitate clinical implementation. Therefore, reporting guidelines have been developed to standardise and streamline the development and validation of AI technologies in health. This commentary proposes a structured approach to utilise these reporting guidelines for the translation of promising AI techniques from research and development into clinical translation, and eventual widespread implementation from bench to bedside. Keywords: Artificial intelligence, clinical translation, digital innovation, guidelines

Presence of microvascular invasion (MVI) indicates poorer prognosis post-curative resection of hepatocellular carcinoma (HCC), with an increased chance of tumour recurrence. By present standards, MVI can only be diagnosed post-operatively on histopathology. Texture analysis potentially allows identification of patients who are considered 'high risk' through analysis of pre-operative magnetic resonance imaging (MRI) studies. This will allow for better patient selection, improved individualised therapy (such as extended surgical margins or adjuvant therapy) and pre-operative prognostication.This study aims to evaluate the accuracy of texture analysis on pre-operative MRI in predicting MVI in HCC.Retrospective review of patients with new cases of HCC who underwent hepatectomy between 2007 and 2015 was performed. Exclusion criteria: No pre-operative MRI, significant movement artefacts, loss-to-follow-up, ruptured HCCs, previous hepatectomy and adjuvant therapy. Fifty patients were divided into MVI (Method 5 achieved the highest accuracy of 87.8% with sensitivity of 73% and specificity of 94%.Texture analysis of tumours on pre-operative MRI can predict presence of MVI in HCC with accuracies of up to 87.8% and can potentially impact clinical management.

Global increase in obesity has led to alarming rise in co-morbidities leading to deteriorated quality of life. Obesity phenotyping benefits profiling and management of the condition but warrants accurate quantification of fat compartments. Manual quantification MR scans are time consuming and laborious. Hence, many studies rely on semi/automatic methods for quantification of abdominal fat compartments. We propose a MultiRes-Attention U-Net with hybrid loss function for segmentation of different abdominal fata compartments namely (i) Superficial subcutaneous adipose tissue (SSAT), (ii) Deep subcutaneous adipose tissue (DSAT), and (iii) Visceral adipose tissue (VAT) using abdominal MR scans. MultiRes block, ResAtt-Path, and attention gates can handle shape, scale, and heterogeneity in the data. Dataset involved MR scans from 190 community-dwelling older adults (mainly Chinese, 69.5% females) with mean age—67.85 ± 7.90 years), BMI 23.75 ± 3.65 kg/m2. Twenty-six datasets were manually segmented to generate the ground truth. Data augmentations were performed using MR data acquisition variations. Training and validation were performed on 105 datasets, while testing was conducted on 25 datasets. Median Dice scores were 0.97 for SSAT & DSAT and 0.96 for VAT, and mean Hausdorff distance was

Pneumonia is one of the most common treatable causes of death, and early diagnosis allows for early intervention. Automated diagnosis of pneumonia can therefore improve outcomes. However, it is challenging to develop high-performance deep learning models due to the lack of well-annotated data for training. This paper proposes a novel method, called Deep Supervised Domain Adaptation (DSDA), to automatically diagnose pneumonia from chest X-ray images. Specifically, we propose to transfer the knowledge from a publicly available large-scale source dataset (ChestX-ray14) to a well-annotated but small-scale target dataset (the TTSH dataset). DSDA aligns the distributions of the source domain and the target domain according to the underlying semantics of the training samples. It includes two task-specific sub-networks for the source domain and the target domain, respectively. These two sub-networks share the feature extraction layers and are trained in an end-to-end manner. Unlike most existing domain adaptation approaches that perform the same tasks in the source domain and the target domain, we attempt to transfer the knowledge from a multi-label classification task in the source domain to a binary classification task in the target domain. To evaluate the effectiveness of our method, we compare it with several existing peer methods. The experimental results show that our method can achieve promising performance for automated pneumonia diagnosis.

Cholangiocarcinoma (CCA) is the second most common primary hepatobiliary malignancy and presents as three separate morphological subtypes; namely mass-forming, periductal-infiltrating, and intraductal-growing patterns. Each of these subtypes have distinct imaging characteristics, as well as a variety of benign and malignant mimics, making accurate diagnosis of CCA on imaging challenging. Whilst histopathological examination is required to arrive at a definitive diagnosis, it is still important for radiologists to be cognizant of these entities and provide reasonable differential diagnoses, as these potentially have a large impact on patient management. This pictorial essay illustrates the three morphological subtypes of CCA, as well as some important mimics for each subtype, that are encountered in clinical practice.

There is increasing appreciation of the association of obesity beyond co-morbidities, such as cancers, Type 2 diabetes, hypertension, and stroke to also impact upon the muscle to give rise to sarcopenic obesity. Phenotypic knowledge of obesity is crucial for profiling and management of obesity, as different fat—subcutaneous adipose tissue depots (SAT) and visceral adipose tissue depots (VAT) have various degrees of influence on metabolic syndrome and morbidities. Manual segmentation is time consuming and laborious. Study focuses on the development of a deep learning-based, complete data processing pipeline for MRI-based fat analysis, for large cohort studies which include (1) data augmentation and preprocessing (2) model zoo (3) visualization dashboard, and (4) correction tool, for automated quantification of fat compartments SAT and VAT. Our sample comprised 190 healthy community-dwelling older adults from the Geri-LABS study with mean age of 67.85 ± 7.90 years, BMI 23.75 ± 3.65 kg/m2, 132 (69.5%) female, and mainly Chinese ethnicity. 3D-modified Dixon T1-weighted gradient-echo MR images were acquired. Residual global aggregation-based 3D U-Net (RGA-U-Net) and standard 3D U-Net were trained to segment SAT, VAT, superficial and deep subcutaneous adipose tissue depots (SSAT and DSAT). Manual segmentation from 26 subjects was used as ground truth during training. Data augmentations, random bias, noise and ghosting were carried out to increase the number of training datasets to 130. Segmentation accuracy was evaluated using Dice and Hausdorff metrics. The accuracy of segmentation was SSAT:0.92, DSAT:0.88 and VAT:0.9. Average Hausdorff distance was less than 5 mm. Automated segmentation significantly correlated R2 > 0.99 (p

Focal liver lesions are commonly encountered. Grey-scale and Doppler sonographic characteristics of focal liver lesions are often non-specific and insufficient to conclusively characterise lesions as benign or malignant. Contrast-enhanced ultrasound is useful for the characterisation of FLLs in patients who are unable to undergo contrast-enhanced computed tomography or magnetic resonance imaging. It is also easily available and relatively cheap. However, interpretation of contrast-enhanced ultrasound can be challenging without a systematic approach. In this pictorial essay, we highlight an algorithm-based approach to FLLs and discuss the characteristic contrast-enhanced ultrasound features of commonly encountered and clinically significant focal liver lesions.

Objectives The 9th International Forum for Liver Magnetic Resonance Imaging (MRI) was held in Singapore in September 2019, bringing together radiologists and allied specialists to discuss the latest developments in and formulate consensus statements for liver MRI, including the applications of gadoxetic acid–enhanced imaging. Methods As at previous Liver Forums, the meeting was held over 2 days. Presentations by the faculty on days 1 and 2 and breakout group discussions on day 1 were followed by delegate voting on consensus statements presented on day 2. Presentations and discussions centered on two main meeting themes relating to the use of gadoxetic acid–enhanced MRI in primary liver cancer and metastatic liver disease. Results and conclusions Gadoxetic acid–enhanced MRI offers the ability to monitor response to systemic therapy and to assist in pre-surgical/pre-interventional planning in liver metastases. In hepatocellular carcinoma, gadoxetic acid–enhanced MRI provides precise staging information for accurate treatment decision-making and follow-up post therapy. Gadoxetic acid–enhanced MRI also has potential, currently investigational, indications for the functional assessment of the liver and the biliary system. Additional voting sessions at the Liver Forum debated the role of multidisciplinary care in the management of patients with liver disease, evidence to support the use of abbreviated imaging protocols, and the importance of standardizing nomenclature in international guidelines in order to increase the sharing of scientific data and improve the communication between centers. Key Points • Gadoxetic acid–enhanced MRI is the preferred imaging method for pre-surgical or pre-interventional planning for liver metastases after systemic therapy. • Gadoxetic acid–enhanced MRI provides accurate staging of HCC before and after treatment with locoregional/biologic therapies. • Abbreviated protocols for gadoxetic acid–enhanced MRI offer potential time and cost savings, but more evidence is necessary. The use of gadoxetic acid–enhanced MRI for the assessment of liver and biliary function is under active investigation.

Splenosis is defined as the autotransplantation of viable splenic tissue throughout various anatomic compartments. Intrahepatic splenosis (IHS) is rare and diagnosis is often challenging. This study aims to provide a comprehensive review on IHS. A literature review was performed on PubMed database. Fifty-six articles with 59 reported cases were included. The majority of the patients were male (n = 49, 83.1%). Median age was 51 years. Risk factors for hepatocellular carcinoma (HCC) included hepatitis B (n = 8, 13.6%) and cirrhosis (n = 12, 20.3%). The majority of the patients were asymptomatic (62.7%) and did not have risk factors for HCC (55.9%). We report a diagnostic triad for IHS: 1) previous history of abdominal trauma or splenectomy, 2) absence of risk factors for liver malignancy and 3) typical imaging features. Non-invasive diagnostic tests such as technetium-99m-tagged heat-damaged red blood cell scintigraphy are useful in diagnosis. Malignancy should be ruled out in the presence of risk factors for HCC.

Targeted biopsy using multiparametric magnetic resonance imaging increases the detection rate of clinically significant prostate cancer (csCaP). In this meta-analysis, we compare the diagnostic accuracy of transrectal (TR) vs transperineal (TP) approaches for MRI-guided software fusion biopsy (FB) in the detection of csCaP. A literature search was performed in PubMed, Cochrane and Embase electronic databases up until July 2019 following the preferred reporting items for systematic review and meta-analysis system. The pooled sensitivity and specificity of either approach was evaluated using radical prostatectomy or systematic biopsies with ≥24 biopsy cores to be the reference standard. Fourteen papers with a total of 2002 patients were selected. Seven hundred and sixty-five patients underwent TR FB, while 1,387 underwent TP FB. One hundred and fifty of the patients underwent both TR and TP approaches. Both approaches were similar in terms of sensitivity (TR vs. TP: 0.81 vs 0.80) and specificity (TR vs. TP: 0.99 vs 0.95). In terms of likelihood ratios and diagnostic odds ratio, TR performed better than TP approach. The area under the receiving operator curve for both approaches was similar (0.91 vs 0.88 respectively). However, there was substantial heterogeneity across the studies for both approaches. TP and TR approaches to software-based FB yield similar diagnostic performance for the detection of csCaP. When deciding on the approach, physicians should consider other inherent features of either technique that suit their practice.

Coronavirus disease 2019 (COVID-19) is caused by the severe acute respiratory syndrome coronavirus 2 and was declared a global pandemic by the World Health Organization on 11 March 2020. A definitive diagnosis of COVID-19 is made after a positive result is obtained on reverse transcription-polymerase chain reaction assay. In Singapore, rigorous contact tracing was practised to contain the spread of the virus. Nasal swabs and chest radiographs (CXR) were also taken from individuals who were suspected to be infected by COVID-19 upon their arrival at a centralised screening centre. From our experience, about 40% of patients who tested positive for COVID-19 had initial CXR that appeared “normal”. In this case series, we described the temporal evolution of COVID-19 in patients with an initial “normal” CXR. Since CXR has limited sensitivity and specificity in COVID-19, it is not suitable as a first-line diagnostic tool. However, when CXR changes become unequivocally abnormal, close monitoring is recommended to manage potentially severe COVID-19 pneumonia. Key words: Diagnostic Radiology, Infectious Diseases, Pulmonary

Objectives: To perform a systematic review and meta-analysis comparing diagnostic performance and inter reader agreement between PI-RADS v. 2.1 and PI-RADS v. 2 in the detection of clinically significant prostate cancer (csPCa). Methods: A systematic review was performed, searching the major biomedical databases (Medline, Embase, Scopus), using the keywords “PIRADS 2.1” or “PI RADS 2.1” or “PI-RADS 2.1”. Studies reporting on head-to-head diagnostic comparison between PI-RADS v. 2.1 and v. 2 were included. Pooled sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were compared between PI-RADS v. 2.1 and v. 2. Summary receiver operator characteristic graphs were plotted. Analysis was performed for whole gland, and pre-planned subgroup analysis was performed by tumour location (whole gland vs transition zone (TZ)), high b-value DWI (b-value ≥1400 s/mm2), and reader experience (Results: Eight studies (1836 patients, 1921 lesions) were included. Pooled specificity for PI-RADS v. 2.1 was significantly lower than PI-RADS v. 2 for whole gland (0.62 vs 0.66, p = 0.02). Pooled sensitivities, PPVs and NPVs were not significantly different (p = 0.17, 0.31, 0.41). Pooled specificity for PI-RADS v. 2.1 was significantly lower than PI-RADS v. 2 for TZ only (0.67 vs 0.72, p = 0.01). Pooled sensitivities, PPVs and NPVs were not significantly different (p = 0.06, 0.36, 0.17). Amongst studies utilising diffusion-weighted imaging with highest b-value of ≥1400 s/mm2, pooled sensitivities, specificities, PPVs and NPVs were not significantly different (p = 0.52, 0.4, 0.5, 0.47). There were no significant differences in pooled sensitivities, specificities, PPVs and NPVs between PI-RADS v. 2.1 and PI-RADS v. 2 for less-experienced readers (p = 0.65, 0.37, 0.65, 0.81) and for more experienced readers (p = 0.57, 0.90, 0.91, 0.65). For PI-RADS v. 2.1 alone, there were no significant differences in pooled sensitivity, specificity, PPV and NPV between less and more experienced readers (p = 0.38, 0.70, 1, 0.48). Inter-reader agreement was moderate to substantial for both PI-RADS v. 2.1 and v. 2. There were no significant differences between pooled csPCa rates between PI-RADS v. 2.1 and v. 2 for PI-RADS 1–2 lesions (6.6% vs 7.3%, p = 0.53), or PI-RADS 3 lesions (24.1% vs 26.8%, p = 0.28). Conclusions: Diagnostic performance and inter-reader agreement for PI-RADS v. 2.1 is comparable to PI-RADS v. 2, however the significantly lower specificity of PI-RADS v. 2.1 may result in increased number of unnecessary biopsies. Advances in knowledge: 1. Compared to PI-RADS v. 2, PI-RADS v. 2.1 has a non-significantly higher sensitivity but a significantly lower specificity for detection of clinically significant prostate cancer. 2. PI-RADS v. 2.1 could potentially result in considerable increase in number of negative targeted biopsy rates for PI-RADS 3 lesions, which could have been potentially avoided.

Advances in our understanding of the role of magnetic resonance imaging (MRI) for the detection of prostate cancer have enabled its integration into clinical routines in the past two decades. The Prostate Imaging Reporting and Data System (PI-RADS) is an established imaging-based scoring system that scores the probability of clinically significant prostate cancer on MRI to guide management. Image fusion technology allows one to combine the superior soft tissue contrast resolution of MRI, with real-time anatomical depiction using ultrasound or computed tomography. This allows the accurate mapping of prostate cancer for targeted biopsy and treatment. Machine learning provides vast opportunities for automated organ and lesion depiction that could increase the reproducibility of PI-RADS categorisation, and improve co-registration across imaging modalities to enhance diagnostic and treatment methods that can then be individualised based on clinical risk of malignancy. In this article, we provide a comprehensive and contemporary review of advancements, and share insights into new opportunities in this field. Published version

Pneumonia can be difficult to diagnose since its symptoms are too variable, and the radiographic signs are often very similar to those seen in other illnesses such as a cold or influenza. Deep neural networks have shown promising performance in automated pneumonia diagnosis using chest X-ray radiography, allowing mass screening and early intervention to reduce the severe cases and death toll. However, they usually require many well-labelled chest X-ray images for training to achieve high diagnostic accuracy. To reduce the need for training data and annotation resources, we propose a novel method called Contrastive Domain Adaptation with Consistency Match (CDACM). It transfers the knowledge from different but relevant datasets to the unlabelled small-size target dataset and improves the semantic quality of the learnt representations. Specifically, we design a conditional domain adversarial network to exploit discriminative information conveyed in the predictions to mitigate the domain gap between the source and target datasets. Furthermore, due to the small scale of the target dataset, we construct a feature cloud for each target sample and leverage contrastive learning to extract more discriminative features. Lastly, we propose adaptive feature cloud expansion to push the decision boundary to a low-density area. Unlike most existing transfer learning methods that aim only to mitigate the domain gap, our method instead simultaneously considers the domain gap and the data deficiency problem of the target dataset. The conditional domain adaptation and the feature cloud generation of our method are learning jointly to extract discriminative features in an end-to-end manner. Besides, the adaptive feature cloud expansion improves the model's generalisation ability in the target domain. Extensive experiments on pneumonia and COVID-19 diagnosis tasks demonstrate that our method outperforms several state-of-the-art unsupervised domain adaptation approaches, which verifies the effectiveness of CDACM for automated pneumonia diagnosis using chest X-ray imaging.

Introduction: Ultrasound (US) is current standard of care for imaging surveillance in patients at risk for hepatocellular carcinoma (HCC). Magnetic resonance imaging (MRI) has been explored as an alternative, given the higher sensitivity of MRI, although this comes at a higher cost. We performed a cost-effective analysis comparing US and a dual-sequence non-contrast MRI (NCEMRI) for HCC surveillance, in the local setting. Methods: Cost-effectiveness analysis of no surveillance, US surveillance and NCEMRI surveillance was performed using Markov modelling and microsimulation. At-risk patient cohort was simulated and followed-up for 40 years to estimate their disease status, direct medical costs, and effectiveness. Quality-adjusted life years (QALYs) and incremental cost effectiveness ratio were calculated. Results: 482,000 patients with an average age of 40 years were simulated and followed up for 40 years. The average total costs and QALYs for the three scenarios – no surveillance, US surveillance and NCEMRI surveillance were S$1,193/7.460 QALYs; S$8,099/11.195 QALYs; S$9,720/11.366 QALYs, respectively. Conclusion: Despite NCEMRI having a superior diagnostic accuracy, it is a less cost-effective strategy than US for HCC surveillance in the general at-risk population. Future local cost-effectiveness analyses should include stratifying surveillance methods with a variety of imaging techniques (US, NCEMRI, CEMRI) based on patients’ risk profiles.

The spleen is a complex organ involved in multiple physiological processes in the human body. Elective splenectomy is an uncommon operation, and the precise characterization of the lesion should be achieved to determine the risks and benefits of this operation accurately. Given the significant role of the spleen in homeostasis and the potential risks of the surgery itself and following sequelae such as infection susceptibility, accurate recognition, and classification of splenic lesions is required before surgery. This review provides an overview of malignant (e.g., lymphoma, angiosarcoma) and benign (e.g., cysts, hemangioma, hamartoma) splenic lesions that may warrant an elective splenectomy. Images from a cohort of adult patients undergoing isolated splenectomy for non-traumatic indications in a single center are provided. This review highlights the considerable overlap in imaging patterns between splenic lesions, splenic lesions masquerading as lesions in other organs, increased detection of asymptomatic splenic incidentalomas due to improvements in imaging modalities. This review also provides clinical correlations for each lesion, providing additional information to help clinicians differentiate between lesions and accurately identify diseases amenable to surgical management.

Gadoxetate magnetic resonance imaging (MRI) is widely used in clinical practice for liver imaging. For optimal use, we must understand both its advantages and limitations. This article is the outcome of an online advisory board meeting and subsequent discussions by a multidisciplinary group of experts on liver diseases across the Asia-Pacific region, first held on September 28, 2020. Here, we review the technical considerations for the use of gadoxetate, its current role in the management of patients with hepatocellular carcinoma (HCC), and its relevance in consensus guidelines for HCC imaging diagnosis. In the latter part of this review, we examine recent evidence evaluating the impact of gadoxetate on clinical outcomes on a continuum from diagnosis to treatment decision-making and follow-up. In conclusion, we outline the potential future roles of gadoxetate MRI based on an evolving understanding of the clinical utility of this contrast agent in the management of patients at risk of, or with, HCC.

Background The presence of concomitant sarcopenia and obesity in sarcopenic obesity (SO) confers worse functional, morbidity and mortality outcomes compared to either alone. Excess adiposity and central redistribution of fats are associated with systemic inflammation and ectopic tissue fat infiltration in forms of Intermuscular adipose tissue (IMAT). Our study examines the profile of IMAT across a spectrum of body compositions and associations with physical performance and inflammatory biomarkers including Monocyte Chemoattractant Protein-1 (MCP-1), a novel biomarker of adipose tissue inflammation. Methods 187 community dwelling elderly participants were recruited and classified into 4 subgroups: normal, obese, sarcopenia and SO, using validated criteria for sarcopenia and waist circumference to define central obesity. We performed magnetic resonance imaging of mid-thigh sections to segment IMAT and muscle. Participants were assessed for muscle strength, physical performance and blood inflammatory biomarkers of interleukin-6, C-Reactive Protein and MCP-1. We examined correlation of IMAT(ratio) with muscle function measures and blood biomarkers. Multiple regression analyses were used to examine the association of body composition types and IMAT(ratio) with muscle function. Results IMAT(ratio) was highest in SO and obese groups. Overall, higher IMAT(ratio) is significantly associated with raised MCP-1, lower gait speed and muscle strength. SO had lowest scores in Short Physical Performance Battery (SPPB), gait speed, hand-grip and knee extension strength. IMAT(ratio) is independently associated with SPPB and handgrip strength, whilst SO is independently associated with muscle strength. Conclusion Our results suggest the possible role of IMAT as a candidate imaging biomarker for adipose tissue inflammation and associated poorer functional outcomes in SO.

Artificial intelligence (AI) has been positioned as being the most important recent advancement in radiology, if not the most potentially disruptive. Singapore radiologists have been quick to embrace this technology as part of the natural progression of the discipline toward a vision of how clinical medicine, empowered by technology, can achieve our national healthcare objectives of delivering value-based and patient-centric care. In this article, we consider 3 core questions relating to AI in radiology, and review the barriers to the widespread adoption of AI in radiology. We propose solutions and describe a “Centaur” model as a promising avenue for enabling the interfacing between AI and radiologists. Finally, we introduce The Radiological AI, Data Science and Imaging Informatics (RADII) subsection of the Singapore Radiological Society. RADII is an enabling body, which together with key technological and institutional stakeholders, will champion research, development and evaluation of AI for radiology applications. Key words: Diagnostic radiology, Machine learning, Neural networks