The Living Laboratory

SOXG Expression in Colorectal Adenomas Improves Surveillance Colonoscopy Risk Stratification in a Bowel Screening Population

Sara Samir Foad Al-Badran¹, Christopher Bigley¹, Mark Johnstone^1,2, Aula Ammar¹, Alexander Winton¹, Jennifer Hay^3,4, Jean Quinn¹, Jakub Jawny⁵, Ditte Andersen⁶, Natalie Fisher⁷, Philip D. Dunne^7,8, Noori Maka5⁹, Gerard Lynch¹, Stephen McSorley^1,5, and Joanne Edwards¹ on behalf of the INCISE Collaborative

Abstract

Adenomas are precursors to colorectal cancer (CRC). Current UK surveillance guidelines use polyp size, number and histology to stratify patients at risk of metachronous polyps/CRC. However, these guidelines are poor at predicting risk, often leading to under/oversurveillance. Adenomas removed from 1257 patients at index colonoscopy were retrospectively used to investigate mutational profile and protein expression trends associated with detection of metachronous polyps/ CRC. The presence or absence of metachronous polyps/CRC was recorded 0.5-6 years after index polypectomy. APC and KRAS were mutated in 87% and 34% of patients, respectively. Both significantly co-occurred with SOX9 mutations (APC 17% p=0.047 and KRAS 23% p=0.012). High cytoplasmic SOX9 expression significantly associated with detection of metachronous polyps/CRC (HR 1.543, p=0.001) and improved risk stratification when combined with current guidelines (HR 2.626, p<0.0001). High cytoplasmic SOX9 expression alone and in combination with current guidelines was an independent predictor of metachronous polyps/CRC according to various regression models. Validation in an independent test dataset confirmed that high cytoplasmic SOX9 expression significantly associated with detection of metachronous polyps/CRC (HR 1.654, p=0.012) and enhanced risk stratification when combined with current guidelines (HR 2.473, p=0.0018).

Authors’ Affiliations

• Wolfson-Wohl Cancer Research Center, School of Cancer Sciences, University of Glasgow, Garscube Estate, Switchback Road, Bearsden, G61 1QH
• School of Medicine, Wolfson Medical School Building, University of Glasgow, University Avenue, Glasgow G12 8QQ
• Glasgow Tissue Research Facility, Queen Elizabeth University Hospital, 1345 Govan Road, Glasgow, G51 4TF
• The Francis Crick Institute, 1 Midland Road, London, NW1 1AT
• NHS Greater Glasgow & Clyde, 1055 Great Western Road, Glasgow, G12 0XH
• BioClavis LTD, Clydebank, 201 Dumbarton Road, Glasgow, G81 4XJ
• Queen’s University Belfast, University Road, Belfast, BT7 1NN
• CRUK Beatson Institute, Garscube Estate, Switchback Road, Bearsden, G61 1BD
• Queen Elizabeth University Hospital, 1345 Govan Road, Glasgow, G51 4TF

Machine Learning Prediction Models of CPET as a Surrogate of Mortality in CHD Patients

Muhammet Alkan*         Gruschen Veldtman†         Fani Deligianni*

* School of Computing Science at University of Glasgow, Glasgow, Scotland, UK

† Golden Jubilee National Hospital, Glasgow, Scotland, UK

Cardiopulmonary exercise testing (CPET) is a global test to assess a patient’s exercise capacity, measuring variables such as oxygen consumption (VO2), carbon dioxide production (VCO2) and pulmonary ventilation (VE) during the exercise. In this study, we aim to predict mortality using a surrogate outcome, VO2 and VE/VCO2 during CPET, which is an independent indicator of high mortality risk in patients with chronic heart failure. We have developed a machine learning approach to fuse information derived from electrocardiograms (ECGs) and clinical letters. Firstly, we started extracting information from clinical letters and stored all patient-related unstructured information such as intervention, diagnosis and medication history in a database using natural language processing techniques. Then, we digitised ECGs to obtain waveforms and linked all the data together, for further analysis with machine learning. This led to 436 patients with congenital heart disease (CHD) being included in our investigation. Combining important clinical notes with the ECG signals, we used these surrogate values as a target variable for our regression model. We exploit the Riemannian geometry of the 12-lead ECG covariance matrices to extract more coherent features. We evaluated our methodology with ablation studies, which showed that integrating ECGs and clinical data yielded the best results.

Immunohistochemical evaluation of Ki67, combined with pathway analysis in pre-malignant colonic polyps, provides insights into potential prognostic marker for metachronous polyps

Aula Ammar¹, Mark Johnstone², Jakub Jawny³, Gerard Lynch¹, Jennifer Hay³, Noori Maka⁴, Stephen McSorley², Joanne Edwards^1*

1-Translational Pathology, Institute of Cancer Science, University of Glasgow; 2- Academic Unit of Surgery, School of Medicine, University of Glasgow, 3- Glasgow Tissue Research Facility, University of Glasgow, 4-NHS Greater Glasgow and Clyde. * On behalf of INCISE collaborative

Bowel screening program risk stratification tool depends on number and size of polyps to assess the need for surveillance colonoscopy following polypectomy. The prognostic significance of index polyps Ki67 for metachronous disease and molecular differences between normal and dysplastic epithelia were explored.

Whole tissue index polyps (n=153) were stained for Ki67 using immunohistochemistry. 48.4% developed metachronous polyps (median time 36.5 months (2006-2019)) and 63% had BSG20 high risk score. QuPath software was used to quantify Ki67. Temp-O-Seq RNA sequencing for normal and dysplastic epithelium (n=40) and ssGSEA/GSEA methods were applied.

Ki67 percentages were higher in luminal (46.4%) vs basal (24.8%) epithelium (p<0.0001).  High luminal Ki67 significantly associated with metachronous disease (p=0.013) and polyp histology (p=0.051) and was an independent prognostic factor for metachronous disease in Cox-regression analysis (HR=2.6, CI(1.238-5.46), p=0.012). Myc targets-V1, G2M checkpoint and E2F-targets were enriched in dysplastic epithelium whereas epithelial-mesenchymal transition was mostly enriched in normal epithelium. Ki67 gene expression was higher in dysplastic than in normal epithelium (p<0.001).

Ki67 is an independent prognostic marker for metachronous polyps and may help improve the current BSG20 criteria.

A third of cancer related deaths globally are attributed to Colorectal cancer. Currently there are limited treatment options available for the 40% of patients who present with a KRAS mutation. At present Regorafenib is the only FDA approved drug for this cohort, but most patients report significant side effects and ultimately discontinue treatment. On the other hand, clean inhibitors such as Trametinib—a MEK inhibitor— fail to provide clinical benefit when used alone. Despite advances in drug development and screening platforms, most compounds fail to translate into clinical benefit. To begin to understand why, my project focuses on introducing a component of the immune tumour microenvironment—macrophages—to capture some of the complexity we must consider in therapeutics. We show that this dynamic cell type can induce resistance to Trametinib, but not to Regorafenib, in both physical and secretory contexts. Within the secretory context, we have identified some key candidates impacting drug response. Our 3D KRAS mutant organoid platform provides an opportunity to explore crucial signalling pathways responsible for this resistance, explore combinations of FDA approved drugs, and suggest key targets for future poly-pharmacological work.

Abstract

Acute myeloid leukaemia (AML) is a heterogeneous blood cancer that predominantly affects the elderly population and confers a poor prognosis. Precision medicine approaches have revolutionised the management and treatment of AML patients, with the development of targeted therapies against numerous disease-driving mutations significantly improving patient outcomes [1]. However, for AML patients with a mutation in the tumour suppressor protein p53 (TP53) gene, there are no effective treatments available, and they have the worst prognosis (2-year survival rate of 13%) [2]. Drugs that target MDM2 and MDMX effectively protect p53 from proteolytic degradation and exploit its tumour suppressive properties in wild-type TP53 cancers but TP53 mutations mediate resistance to these compounds [3]. Cancers lacking functional p53 retain dependency on MDM2/MDMX activity [4] and identifying superior approaches to block MDM2/MDMX activity beyond their p53-dependent roles could have promise for the treatment of TP53-altered AML. Here, we show that a novel cell-penetrating peptide (DRx-992) that inhibits MDM2/MDMX activity through dimerisation disruption demonstrates cytotoxic activity against a heterogenous panel of AML cell lines and outperforms clinical stage MDM2 inhibitor idasanutlin in TP53 mutant cells. The broad efficacy observed suggests the peptide may have the potential to treat diverse AML subtypes, including TP53 mutant AML.

References

[1] Döhner H, Wei AH, Löwenberg B. Towards precision medicine for AML. Nature Reviews Clinical Oncology, 2021;18:9 577–590.

[2] Grob T, Al Hinai ASA, Sanders MA, Kavelaars FG, Rijken M, Gradowska PL et al. Molecular characterization of mutant TP53 acute myeloid leukemia and high-risk myelodysplastic syndrome. Blood, 2022; 139: 2347–2354.

[3] Garnett MJ, Edelman EJ, Heidorn SJ, Greenman CD, Dastur A, Lau KW et al. Systematic identification of genomic markers of drug sensitivity in cancer cells. Nature, 2012; 483: 570–575.

[4] Klein AM, De Queiroz RM, Venkatesh D, Prives C. The roles and regulation of MDM2 and MDMX: It is not just about p53. Genes Dev, 2021; 35: 575–601.

Introduction

Clinical gait analysis (CGA) plays a vital role in diagnosing and understanding rehabilitation of movement disorders by analysing joint kinematics, such as flexion/extension angles. Conventional marker-based motion capture (mocap) systems have been long considered the gold standard for kinematic measurements in CGA. However, markerless systems such as Theia3D offer a promising alternative which may enhance clinical usability by eliminating need for markers.

Objective

This study aims to compare the accuracy of flexion/extension angles obtained from Theia3D to a marker-based system, for treadmill walking.

Methods

26 participants (12 female, 14 male) completed a 60-second self-paced treadmill walk while being simultaneously recorded by both marker-based and Theia3D mocap systems, each sampling at 100Hz. Joint flexion/extension angles were processed using Visual3D, with subsequent root mean square error (RMSE) calculated.

Results

The comparison of flexion/extension angles showed mean RMSE values of 4.97° for the ankle, 3.38° for the hip, and 5.30° for the knee across participants.

Conclusion

Theia (v2023) showed comparable accuracy to marker-based motion capture for flexion/extension angles, with varying RMSE across joints. These results suggest that Theia offers a feasible alternative for CGA, particularly where marker-based limitations may be a concern. Further research may clarify its application across different clinical settings.

In-stent restenosis (ISR) poses a risk for patients who have undergone stent implantation following atherosclerosis. It consists of the gradual intimal growth that follows the stenting of a patient potentially leading to the vessel’s new occlusion.

Computational models have been implemented in a continuous and discrete form to study ISR, incorporating various factors concurring to cause it and offering predictions on the procedure’s outcomes. This work aims at presenting a patient-specific multiscale model of an atherosclerotic coronary artery including different frameworks: a structural finite element module (FEM) integrated in a drug transport module and an agent-based module (ABM). The stent deployment module generates a deformed geometry. The drug transport module takes as input this geometry and computes the drug retention and distribution in the tissue. Finally, the ABM module simulates cell and extracellular matrix dynamics to obtain the tissue growth and arterial remodelling.

The proposed work successfully reproduced how the intimal tissue regrowth is influenced by the presence of a drug released over time and inflammation. This multiscale model represents an innovative tool for the studying of ISR by presenting a cellular and tissue perspective on a patient-specific artery characterized by an atherosclerotic plaque with regions of different nature.

Title: Autonomous Robotic System for Precise Placement of Dental Implants in Edentulous Jaw Models

Abstract: This poster presents our recent progress on developing a novel autonomous robotic system for dental implants in edentulous jaw models. The system integrates a robotic arm with six Degrees of Freedom (DoF) with a dental navigation system. The hand-eye calibration is applied to synchronize the robotic arm with the navigation system through a precise transformation matrix. Key engineering innovations include the development of wireless transmission for real-time coordination between the navigation system and the robotic arm, alongside automated depth control for the implant placement process. The implants were placed, with 3D deviations of 1.80 mm at entry and 2.80 mm at exit points, within clinically acceptable limits. These results demonstrate the potential for this system to enhance precision and consistency in dental implant procedures, reducing operator-dependent variability.

Abstract

The exponential growth of medical data from electronic health records (EHRs) and clinical sources presents both opportunities and challenges in healthcare analytics. This research proposal aims to develop an innovative method for high-confidence medical data association analysis using knowledge graphs, with a particular focus on leveraging Neo4j, a leading graph database management system.

The study will construct a comprehensive knowledge graph using Neo4j, integrating diverse medical data sources including EHRs, laboratory test results, and established medical knowledge bases. Neo4j's powerful graph data model and query language (Cypher) will be utilized to efficiently represent and traverse complex medical relationships. Novel algorithms will be developed to exploit Neo4j's capabilities for high-confidence association analysis, addressing critical challenges in data standardization, integration, and classification of large-scale patient data.

Key components of the research include data preprocessing, knowledge graph construction in Neo4j, confidence calculation using graph-based metrics, association analysis through optimized graph traversal, and data classification leveraging Neo4j's built-in algorithms. The method will be validated using real-world medical data, with performance compared against traditional relational database approaches through controlled experiments.

Expected outcomes include a validated high-confidence association analysis method built on Neo4j, efficient graph-based algorithms for identifying significant patient associations, and a comprehensive framework for medical data analysis using graph databases. This research has the potential to significantly enhance diagnostic accuracy, treatment planning, and clinical decision support, while demonstrating the advantages of graph databases like Neo4j in handling complex, interconnected medical data.

Clonal haematopoiesis and cardiovascular morbidity in relapsed refractory solid organ malignancy

Samantha Drummond, Stephanie Anderson, Mhairi Copland, Gillian Horne, Patricia Roxburgh

Background

Clonal haematopoiesis (CH) occurs through acquisition of somatic mutations and resultant clonal expansion. It is an independent risk factor for cardiovascular disease (CVD). Patients with solid organ malignancies have a higher prevalence of CH.

In the Tumour Characterisation to Guide Experimental Targeted Therapy (TARGET) National study, next generation sequencing was performed on circulating cell-free DNA. TARGET reports included mutations for targeted oncological treatment and mutations associated with CH. 

Aim

To determine the prevalence of CH and cardiovascular morbidity in the TARGET population within a single centre.

Methods

Patients were identified by review of TARGET reports. Data on mutational status was collected from these reports. Clinical details were collected from electronic patient records.

Results

A total of 85 patients were identified. The median age was 58 years (18-83). Nearly all patients (98.8%) received prior chemotherapy and 29.4% received radiotherapy.

Forty-four patients (51.8%), had a mutation consistent with CH. Although not statistically significant, a greater proportion of patients with CH had CVD, diabetes and hypertension compared to those without CH.

Conclusions/Discussion

The higher prevalence of CH in this project may be due to exposure to oncological therapies. Further studies are required to determine the long-term outcomes of oncology patients with CH.

Virtual Reality Gaming for Upper Limb Rehabilitation

Stroke is a leading cause of death and adult acquired disability worldwide, with about 40-50% of Stroke survivors experiencing impairments in the upper limb (UL). The UK national clinical guidelines for stroke recommends a stroke patient receives at least 45 minutes of that per day for each type of needed therapy. The shortage of therapists in the NHS and diverse rehabilitation needs of patients makes this difficult to achieve, thus patients not receiving the adequate amount of therapy that they need.

With the increased need for interventions, research into the use of Serious Games (SG) and Virtual Reality (VR) interventions to augment rehabilitation therapy is also increasing, with some promising results which show that the use of 3D immersive SG in VR for UL and cognitive rehabilitation increases the personalised nature of rehabilitation and mitigate other challenges associated with rehabilitation experiences e.g. fatigue, monotony, demotivation, and lack of engagement.

The aim of this study was to co-design immersive VR SGs for UL rehabilitation for people with stroke by incorporating both end-users and rehabilitation clinical teams’ insights.

The co-design approach revealed that end-users were open to trying different games for therapy purposes and that therapists also appreciated SG VRs rehabilitation support.

Name: Narinder Kaur 

Department: School of Cardiovascular Science and Metabolic Health

Role: Precision Medicine Scientist

Title: Novel Support Tool for Predicting Incident Heart Failure in Type-2 Diabetes: A Precision Medicine Approach

Introduction

The incidence of heart failure (HF) is higher among individuals with type-2 diabetes (T2D). Symptoms and signs of HF often remain undetected and manifest late in the disease trajectory. Novel machine-learning (ML) tools may help clinicians to identify patients with T2D who are at increased risk of developing HF.

Aim

To develop a novel support tool for heart failure risk assessment in patients with T2D.

Methods

We obtained electronic medical records (EMRs) from two different populations (UK and China) including demographics, laboratory measurements, medications and prior comorbidities. Incident HF was defined using International Classification of Diseases, 9th and 10th Revision (ICD-10) codes. We implemented time-dependent machine-learning models. We integrated state-of-the-art artificial intelligence interpretability with clinical expertise to provide concise reasoning for a patient's increased risk of HF.

Results

Of 29,868 patients in UK with T2DM age >50 years, 976 (3%) had incident HF between 2009-19, The baseline model c-statistic score was 0.87 and the time brier score was 0.02 for predicting incident HF. Of 262,687 patients in China with T2D aged >55 years, 8,515 (3%) had incident HF between 2009-19. The baseline model c-statistic score was 0.88 and the time brier score was 0.07. Important predictors of HF risk were use of loop diuretics, prior atrial fibrillation (AF), coronary artery disease (CAD), Also, older age, lower haemoglobin, lower estimated glomerular function rate (eGFR), lower serum albumin, higher HbA1c and higher alkaline phosphatase (ALP).

Conclusion

Using readily available patient EMRs and machine-learning, the incidence of HF for patients with T2D in both UK and China can be predicted with similar accuracy, suggesting that it is applicable to diverse populations.

Langevin Transducers Incorporating TPMS Lattice Front Masses

Abstract

Bolt-clamped Langevin transducers (BLTs) are widely used in ultrasonic surgical procedures but opportunities to tailor their vibrational responses have been limited. Unlike the common approaches for geometric modifications, new ways of controlling the vibrational response, can be realised through additive manufacturing. We introduce triply periodic minimal surface (TPMS) lattice structures and explore their potential for the development of front masses. A set of FEA simulation and experimental characterisation was performed in this study. It is shown that an evolution of vibrational behaviours occurred in the transducers with different volume and positions. First longitudinal mode converted into longitudinal-bending hybrid mode in transducers as notable enhancements of bending occur from the compliant lattice region. A proper amount of gyroid lattice structures lowered the compliance of front masses and induced a different degree of increase in displacement amplitude and gain values. Cracks alongside surface adhered powder, and relatively low volume of micro-pores were visualised in displacement amplified structures after excitation. TPMS lattice as a first attempt in transducer design showed its potential to achieve displacement amplification and mode coupling. However, a further investigation of structural integrity is necessary to make them ready for medical practices.

Title: ‘ForceBiology: Cellular Forces as Predictors of Clinical Success’

Authors: Maia Lyall MRes, Badri Aekbote PhD, Ross McKeown MSc, Nikolaj Gadegaard PhD

Word Count: 199

Drug discovery is an expansive market, but suffers from low success rate, with >90% of drugs going through the clinical pipeline never making it to the market. This is often due to drug efficacy and toxicity. Cardiac toxicity accounts for 30% of drug failure and abandonment. Current methods of pre-clinical in-vitro cardio-toxicity screening are complex, qualitative, costly and identifying cardiotoxicity too late in drug development, leading to costly failures during clinical trials.

There are no direct tools available to quantify the contractile forces in cells like cardiomyocytes or smooth vascular muscle cells in a reliable, efficient and (HTS) high-throughput manner.

We have developed a HTS drug screening platform: ForceBiology, that uses traction force microscopy to probe cellular mechanics and provide quantitative output for cellular force in contraction and migration.

Single cell force measurements in this platform provide an avenue to select better drugs and make more informed decisions in the drug discovery pipeline.  This technology provides options for personalized drugs, with the aim to reduce expensive failures in later stage clinical development, by improving clinical predictability.

We propose that this platform will improve healthcare strategies by reducing trial-and-error treatment approaches, minimizing costs, time, animal studies and improving patient outcomes.

Engineered Circular Spiral Micro-Coils for Deep Brain Neurostimulation

Tahereh (Tala) Masalehdan (PhD candidate)

Microelectronics Lab (meLAB), James Watt School of Engineering, University of Glasgow, G12 8QQ, UK

Abstract— Neurostimulation techniques are crucial in advancing neuroscience research and addressing neurological disorders. Within this realm, magnetic neurostimulation stands out by offering numerous benefits compared to conventional methods like electrical stimulation. Through the current work the feasibility of employing spiral µ-coils for magnetic neurostimulation was explored both empirically and experimentally. Simulations predicted average magnetic (B) field values ranged from 8.41 to 42.05 mT for currents between 1 and 5 A across all frequencies tested, enabling penetration into the model up to a depth of around 2 cm for neurostimulation. The optimal laser cutting power was identified as 2.6 W and the results demonstrate that the B-field exhibited variability, ranging from 1.72 µT at the coil's center to 772.26 nT at 25 mm from the center. These findings suggest that magnetic neurostimulation holds promise as a viable approach for targeting brain regions located beneath the cortex.

The current shift towards tailored, patient orientated treatment of cancer is nowhere more welcomed than in paediatric cancer. Standard of care chemotherapeutics have significant systemic consequences on children; however, few novel therapies have provided more reputable results. There remains a significant unmet need for targeted therapies, such as Cell Penetrating Peptides (CPP), in the treatment of paediatric cancers, specifically High-Risk Neuroblastoma (HR-NB). HR-NB represents approximately 50% of neuroblastoma cases (1) and can be characterised as a greater prevalence of TP53 mutations that render p53, the “guardian of the genome”(2), incapable of tumour suppression. Furthermore, increased MYCN amplification and elevated Mouse Double Minute 2 (MDM2) and Mouse Double Minute X (MDMX) dependency are also common characterisations (3). Existing similar small molecule therapies for neuroblastoma are unable to exploit the innate protective role of p53 in TP53 mutant cases (4). MDM2/X are negative regulators of p53 and upregulated in many cancers. Their ability to homo- and heterodimerise presents an interesting novel targeted therapy approach. Here we describe a novel cell penetrating MDM2/X dimerization inhibitor (DRx-992), which we hypothesise activates pro-apoptotic function through p53 independent mechanisms, making it effective in both TP53 mutated and wild-type neuroblastoma, contrasting current industry standards (figure 1)(4).

Tri-Sectoral Collaboration in Translation of a Digital Health Technology: Perspectives of Academia, Industry, and Healthcare Provider Partners

Authors: Miss Dikshyanta Rana¹, Dr Janet Bouttell^1,2, Dr Nicola McMeekin¹, Dr Gerard Lynch³, Prof Joanne Edwards³, Prof Neil Hawkins¹

Author affiliation:

1. Health Economics and Health Technology Assessment
   School of Health and Wellbeing
   University of Glasgow
   Glasgow, G12 8TB
   United Kingdom
2. Centre for Healthcare Equipment and Technology Adoption
   Nottingham University Hospitals Trust
   Nottingham, NG7 2UH
   United Kingdom
3. Translational Cancer Pathology Group (Edwards Lab)

School of Cancer Sciences
University of Glasgow
Glasgow, G61 1QH
United Kingdom

Background:

Translating digital health technologies into real-world healthcare settings is a lengthy and highly complex process. Tri-sectoral collaborations involving academia, industry, and healthcare provider (AIHP) partners can play a pivotal role in this journey. This study explores the intricacies of such tri-sectoral collaborations during their initial phases.

Methods:

This qualitative study utilised a large, retrospective research project aimed at developing a risk stratification tool for identifying patients at risk of metachronous polyp development or colorectal cancer. Semi-structured interviews explored topics such as product knowledge, regulatory pathways, evidence requirements, and factors influencing barriers, facilitators, and sustainability. A thematic analysis was performed.

Results: 

Six key themes highlighting the intricacies of AIHP collaborations were identified: a) collaboration dynamics, b) envisioning the final output, c) securing funding throughout the project lifespan and beyond, d) navigating regulatory pathways, e) developing commercialisation strategies, and f) other considerations.

Conclusion:

Tri-sectoral collaborations offer significant advantages in bridging the translation journey. However, they also encounter interrelated, multi-dimensional challenges—not only in the translation processes but also within the collaborative framework itself. These challenges may hinder collaboration success, potentially affecting the technology's innovative value and delaying its timely introduction for patient benefit. Thus, AIHP partners must proactively identify and openly discuss strategies for tackling these challenges from the outset of their collaboration. This study provides valuable insights and lessons for future researchers engaged in similar collaborations.

Title: Understanding the role of inelasticity in Drug-Coated Balloon deployment for improved Percutaneous Coronary Intervention outcomes

Authors:

Silvia Renon (1), Richard Good (2), Ankush Aggarwal (1), Sean McGinty (1)

1. James Watt School of Engineering, University of Glasgow, UK; 2. School of Cardiovascular & Metabolic Health, University of Glasgow, UK.

Abstract:

Drug Coated Balloons (DCB) represent an emerging therapeutic alternative to Drug Eluting Stents (DES) for the treatment of coronary artery disease, particularly for complex lesions. Despite their advantages, such as the absence of permanent metallic structures and efficient drug delivery, the underlying mechanisms of drug elution and the influence of arterial mechanics remain incompletely understood. This study aims to develop an in-silico model to investigate the critical role of inelasticity in the balloon deployment process during Percutaneous Coronary Intervention (PCI).

Utilizing advanced finite element analysis within COMSOL Multiphysics, we constructed a realistic 2D model of atherosclerotic arteries where perfect plasticity and linear hardening inelastic models were tested to simulate the arterial state after balloon expansion. Patient-specific plaque geometries were analysed to evaluate the impact of lesion preparation on drug elution and restenosis risk. Preliminary results indicate that alterations in plaque composition exert influence on the distribution of plastic strains and, consequently, on residual stresses in the arterial wall after balloon expansion.

This research focuses on the understanding of DCB dynamics in relation to arterial mechanics. By comprehending the interactions between balloon mechanics and arterial tissue, this study contributes valuable insights for improving clinical outcomes in PCI.

Title

Rapid and modular workflow for same-day sequencing-based detection of bloodstream infections and antimicrobial resistance determinants

Authors:

Mohammad Saiful Islam Sajib¹, Katarina Oravcova¹, Kirstyn Brunker^1,2, Paul Everest¹, Manuel Fuentes¹, Catherine Wilson³, Michael E. Murphy^3,4, Taya Forde¹

Affiliations:

¹School of Biodiversity, One Health & Veterinary Medicine, University of Glasgow, Glasgow, United Kingdom

²MRC-University of Glasgow Centre for Virus Research, Glasgow, United Kingdom

³Department of Microbiology, NHS Greater Glasgow and Clyde, Glasgow Royal Infirmary, New Lister Building, Alexandra Parade, Glasgow, UK

⁴School of Medicine, Dentistry & Nursing, College of Medical, Veterinary & Life Sciences, Wolfson Medical School Building, University of Glasgow, Glasgow, UK

Abstract

Bloodstream infections (BSI) are a major public-health concern, and existing diagnostic technologies are suboptimal for guiding timely and targeted antibiotic therapy for critically ill patients. Nanopore metagenomic-sequencing (mNGS) can facilitate rapid microbiological diagnosis, but identification is challenged by significant host versus bacterial DNA in blood. We developed M-15, a rapid mNGS-based chemical host DNA depletion (CHDD) workflow, benchmarked with five commercial/published protocols and validated with BACT/ALERT positive/negative samples from BSI cases and rapid culture-enriched spiked blood.

All six CHDD protocols removed 2.5×10⁰ to 4.1×10⁶-fold host DNA, with M-15 performing the best, while adaptive sampling removed host >5-fold. With BACT/ALERT specimens, M-15 mNGS accurately identified 3/3 negative, 28/28 mono-bacterial, and 2/4 multi-bacterial species. With rapid culture-enrichment and M-15 mNGS, <18% DNA was classified as host and all bacterial species tested (n=10) were correctly identified. M-15 mNGS accurately predicted phenotypic antimicrobial resistance (AMR)/susceptibility for 90.3% (232/257) of drug/bacteria combinations from BACT/ALERT positive samples.

This study demonstrates that M-15 mNGS can facilitate species and AMR gene detection within 5-7 hours of BACT/ALERT positivity. Including 8-hour culture enrichment, microbiological and AMR confirmation is possible within 13-15 hours of sample collection. Thus, M-15 mNGS workflow has the potential to improve patient outcomes in BSI.

“Long way off from a human… this is something that can work alongside us”: A longitudinal qualitative evaluation of the acceptability of artificial intelligence software to prioritise chest X-rays

Emma J Steel ¹, Sean F Duncan ², Mark Hall ³, David B Stobo ³, John D Maclay ^{4, 5}, Shamie Kumar ⁶, David J Lowe ^{2, 7}, Evi Germeni ¹

1. Health Economics and Health Technology Assessment (HEHTA), University of Glasgow, UK

2. Digital Health Validation Lab, College of Medical, Veterinary & Life Sciences, University of Glasgow, Glasgow, UK

3. Radiology, Queen Elizabeth University Hospital, Glasgow, UK

4. Respiratory Medicine, Glasgow Royal Infirmary, Glasgow, UK

5. School of Medicine, Dentistry & Nursing, University of Glasgow, Glasgow, UK

6. Qure.AI, Mumbai, India

7. Emergency Department, Queen Elizabeth University Hospital, Glasgow, UK

Background: Plain film chest X-rays (CXRs) remain the first line investigation on the lung cancer diagnostic pathway. The qXR artificial intelligence (AI) software was developed by Qure.ai to assess CXRs and flag urgent suspected cancers as a priority for reporting. A longitudinal qualitative evaluation was designed to provide an in-depth understanding of the acceptability of the software among key stakeholders.

Methods: In-depth interviews were conducted with 16 NHS staff (radiologists, radiographers, administrators) before and 5+ months after implementation of the software. The Theoretical Framework of Acceptability informed data collection; data analysis employed an inductive approach, identifying themes from participants’ expressed perspectives. 

Results: Overall, participants viewed the software as a useful adjunct and believed it could expedite lung cancer pathways. While they were surprised by how easily it integrated into their systems, they also expressed concerns about its performance, including misreporting by the AI, disagreement with its interpretations, biased decision-making, and potential liability issues. Participants recognised the challenges of implementing the software and highlighted the impact on the patient pathway as a key issue.

Conclusions: The successful adoption of AI in radiology requires not just a software that integrates easily, but one which also performs accurately and does not disrupt practice.

Machine learning to simplify complex coronary small vessel diagnostics

Authors: Robert A. Sykes^1,2, Daniel T. Ang^1,2, Dylan Tan², Colin Berry^1,2

Institutional Affiliations: 1 School of Cardiovascular & Metabolic Health, University of Glasgow, Glasgow, UK; 2 West of Scotland Heart and Lung Centre, Golden Jubilee National Hospital, Clydebank, UK.

Presenting author: Dr Robert A. Sykes, Clinical Lecturer in Ischaemic Heart Disease (SCREDS), Specialty Registrar in Cardiology and General Internal Medicine, School of Cardiovascular and Metabolic Health, College of Medical, Veterinary and Life Sciences, University of Glasgow.

Abstract

Background:

This study explores the potential of machine learning and predictive modelling as safer, economically attractive alternatives to invasive coronary physiology for assessing microvascular function.

Methods:

Data from patients with angiographically unobstructed coronary arteries, who underwent invasive coronary thermodilution physiology assessment, were retrospectively analysed. A feature scaled linear regression model predicted coronary transit times using the resting TIMI frame counts at rest. This was combined with a standardised multi-layer perceptron neural network regression (MLPR) model to predict non-linear parameters including the hyperaemic responses for distal coronary artery blood pressure and flow. Cross validation to establish the optimal hyperparameters and prevent overfitting was undertaken for neural network models prior to analysis.

Results:

The MLPR model predicted hyperaemic distal coronary pressure (Pd) from resting proximal aortic pressure (Pa) with an R² of 0.59. TIMI frame count predicted resting coronary transit time with an R² of 0.33, with MLPR addition for the calculation of hyperaemic transit time R²=0.45. The combined model predicted the index of microvascular resistance (IMR) and coronary flow reserve with 97% and 68% of the variance explained, respectively.

Conclusions:

These results indicate that a machine learning model is feasible for the assessment of coronary microvascular function and potentially enhances our understanding of coronary physiology.

Table 1

Singlet Oxygen Luminescence Detection During Photodynamic Therapy for Cancer Treatment

Vikas¹, Weibing Yang², Baozhu Lu², Brian C Wilson³, Timothy C Zhu², Robert H Hadfield¹

1James Watt School of Engineering, University of Glasgow, Glasgow G128LT, UK,

2 Department of Radiation Oncology, University of Pennsylvania, Philadelphia, PA, 19104, USA

3Department of Medical Biophysics, University Health Network/University of Toronto, Toronto, Canada

Photodynamic therapy (PDT) is an emerging cancer therapeutic modality that utilizes light, a photosensitizer, and molecular oxygen interaction to induce localized cell death. This treatment's effectiveness is attributed to the generation of cytotoxic singlet oxygen (1O2) in type II PDT. Accurate monitoring of 1O2 levels during PDT enables targeted treatment delivery to the tumor site while reducing non-specific effects. Direct observation of the 1O2 luminescence at 1270 nm poses a significant challenge due to its low intensity. Time-resolved Singlet Oxygen Luminescence Detection (TSOLD) is a real-time technique for measuring 1O2 during PDT, utilizing a supercontinuum laser, a cuvette with the photosensitizer solution, optical filtering, parabolic mirrors, a single-photon avalanche diode (SPAD) detector, and time-tagger electronics. The 1O2 luminescence signal generated by Protoporphyrin IX (PpIX) in ethanol and acetone solutions is measured using an in-house developed TSOLD system. The concentration-dependent analysis showed that as the PpIX doses within solutions increased from 1 to 10 mg/kg, there was a 3-4x amplification in the 1O2 luminescence signal. The lifetime of 1O2 for PpIX shows significant variability depending on the solvent used, with values of ~14.5 μs in ethanol and 48.3 μs in acetone. This study demonstrated the considerable role of 1O2 luminescence detection in quantifying PDT efficacy.

Self-sterilising Door Handles

The Bio-Interface Group at Glasgow University have developed antimicrobial and photo-activated nanocoatings for touch surfaces and safely integrated them with germicidal UV light to produce smart, self-sterilizing systems. This technology has many applications. In partnership with university start-up, Liteworx Ltd, a smart, self-sterilizing pull handle suitable for doors in communal environments where cross-contamination is most frequent has been developed. This technology is timely and would be an asset for care home infection control and promoting welfare for the hospitality industry in post-COVID times. The prototype is presently of Technology Readiness Level 6, suitable for environmental testing so may be a suitable product for further development with support form The Living Laboratory.