Stratford University

The rising cases of diabetes on the earth have increased the necessity to possess scaled-up eye-care systems that may detect the pathology of retinal illnesses before the permanent loss in sight had occurred. The application of retinal image screening and grading algorithms, made by artificial intelligence, has become an administrative solution to Diabetic Retinopathy (DR) screening and grading of retinal images, especially ones that lack specific expertise. Deep Learning models are currently expert level based on feature abstraction, lesion detection and multi-stage DR classification with the use of large annotated fundus datasets. The directions of the recent past focus on using the hybrid processing pipelines, smaller neural networks, and optimization of training to enhance the diagnostic performance across a range of images scenarios. Clinical-grade labelling, increased contrast normalization, attention-gated reasoning and interpretability, have been designed to make such systems more friendly to ophthalmic-decision-making. In this paper, I have undergone a thorough analysis of AI-assisted DR grading relative to a regular high-parameter residual network against a more modern and parameter-efficient scaled architecture. The given workflow includes synthetic data creation, pre-processing algorithms and alleviating the problem of class imbalance, and stricter evaluation according to the results in terms of accuracy, AUC, F1-score, and analysis of the confusion matrix. Additional information about how the model behaves at realistic screening conditions can be provided by the quantitative comparisons (ROC curves, convergence profiles and confidence distributions). It has been found that scalable artificial intelligence (AI) systems are associated with stable diagnostic, computationally economical, and robust results, indicating their applicability in real practice in resource-intensive clinical environment.

Real time detection of cancer is an essential issue in the contemporary healthcare, especially in resource limited clinical settings and edge devices. Current methods of deep learning, which are mostly based on Transformer architectures, have quadratic computational complexity, which constrains their use on mobile and embedded systems. In this paper, the author proposes a novel model architecture, MambaOELM (Mamba-based Optimized Edge Lightweight Model), which is an efficient, real-time architecture of detecting cancer, using a state-space model. Using the selective state-space mechanisms with linear-time complexity, quantization, and pruning, MambaOELM can attain the inference speeds that can be deployed on the edge even with relatively low-resource costs, in addition to competitive accuracy across a variety of cancer types. We achieve 94.2% accuracy and 23 ms inference latency on edge devices with our experimental evaluation to detect breast cancer, which is a 4.8x improvement in speed over our transformer-based baselines and a 73% model size reduction. The suggested architecture will facilitate the usage of advanced cancer detection systems in resource constrained clinical facilities, mobile health applications and IoT facilitated diagnostic devices.

This study explores the intricate strategic implications of digital transformation for large established organisations operating within dynamic environments. While digital transformation has garnered substantial attention in the context of startups, large established organisations face unique challenges in balancing existing capabilities with adopting new digital capacities. This research sheds light on how large established organisations navigate the interplay between change and stability during digital transformation by employing a dynamic capabilities framework. Through a systematic literature review of 123 articles from leading journals, the research categorises large established organisations dynamic capabilities into four domains: performance, leadership, governance and structure. The findings reveal that for each domain, large established organisations encounter a range of dilemmas during digital transformation. The study highlights the need for large established organisations to address these dilemmas and operationalise digital transformation effectively. By offering a comprehensive perspective on these challenges, the study provides valuable guidance for researchers and managers seeking to navigate the complexities of digital transformation in established organisations. The paper concludes by outlining potential directions for future research in this evolving field.

2667 Background: ICI use is linked to severe gastrointestinal (GI) immune-related adverse events (irAEs), which affect morbidity and mortality and often require treatment pauses. Among these, immune-mediated colitis (IMC)—primarily associated with CTLA-4 therapy—occurs in 5.7% to 39.1% of patients receiving CTLA-4 inhibitors and 0.7% to 31.6% of those receiving PD-1/PD-L1 inhibitors; combination therapy can raise this incidence to 40.4%. IMC symptoms range from mild diarrhea to severe colitis, typically requiring urgent intervention within six to eight weeks of immunotherapy to prevent complications such as colonic perforation or sepsis. Corticosteroids are the usual first-line treatment, with TNF-alpha inhibitors (e.g., infliximab) considered when patients do not improve after three to seven days. Vedolizumab, a gut-selective α4β7 integrin antagonist that targets gastrointestinal-homing T-lymphocytes, offers an alternative approach. Both infliximab and vedolizumab—referred to as Selective Immunosuppressive Therapies (SITs)—have shown promise, though their distinct mechanisms have led to a lack of standardized protocols and reliance on provider discretion. This study compares infliximab, vedolizumab, and combined SITs (infliximab plus vedolizumab) in managing IMC, focusing on remission rates, recurrence, and improved steroid tapering success. Methods: A systematic search was conducted across the PubMed database. The Meta-Analysis was conducted using R version 4.4.1 to calculate odds ratios (ORs) and 95% confidence intervals (CIs). Results: A total of eight studies were included in the final analysis. In patients with immune checkpoint inhibitor–induced colitis, vedolizumab was associated with higher rates of colitis recurrence (OR = 0.32, 95% CI = 0.19–0.53) compared to infliximab. Patients receiving vedolizumab also had lower overall corticosteroid usage (mean difference in days: -18.29, 95% CI = -21.88 to -14.71) compared to infliximab recipients. There was no significant difference in remission rates between vedolizumab and infliximab monotherapy; however, higher remission was noted with combination therapy (vedolizumab plus infliximab) (OR = 0.40, 95% CI = 0.19–0.84) compared to infliximab monotherapy. Conclusions: Vedolizumab was associated with a higher recurrence rate of colitis but resulted in significantly lower corticosteroid usage compared with infliximab. Although remission rates were similar for both monotherapies, combination therapy (vedolizumab plus infliximab) demonstrated higher remission rates than infliximab alone.

FXYD5 is involved in various biological processes, including inflammation, tumor progression, drug resistance, and hypertension. It has been suggested as a potential biomarker for several cancers, with elevated levels found in ovarian and colon cancer. However, the role of FXYD5 in prostate cancer has not yet been elucidated. Here, we found that the FXYD5 was expressed in tumor cells by sc-RNA sequencing of human metastatic prostate tumor tissues. The expression levels of FXYD5 were higher in androgen receptor (AR)-negative prostate cancer cells compared to AR-positive prostate cancer cells. Intrinsically, FXYD5 knockout promoted PC3 cell proliferation, migration, and invasion. In vivo studies showed that FXYD5 knockout promoted PC3 tumor growth, while overexpression of FXYD5 inhibited PC3 tumor progression. However, in AR-positive 22Rv1 cells, overexpression of FXYD5 promoted tumor progression. In vitro assays indicate that overexpression of FXYD5 in PC3, C42B, and 22Rv1 cells reduces mitochondrial membrane potential and reactive oxygen species. Additionally, FXYD5 overexpression reduces lactate dehydrogenase activity and lactate levels in PC3 cells but has no effect on these parameters in 22Rv1 cells. In PC3 cells, FXYD5 overexpression promotes the formation of tight cell-cell interactions, accompanied by downregulation of membrane markers and nuclear enlargement. In contrast, these effects are not observed in 22Rv1 or C42B cells. Further, FXYD5 is identified as a potential Siglec-7 ligand by CRISPRi screen. FXYD5 knockout decreased Siglec-7 Fc binding capacity and enhanced the NK-cell mediated cytotoxicity in prostate cancer cells. In a humanized mouse model, 22Rv1 tumor growth was inhibited with human immune cell modulation. However, when FXYD5 was overexpressed in 22Rv1 cells, tumor growth was not impacted by immune cell-based therapy. This was due to the interaction of FXYD5 on cancer cells with Siglec-7 on immune cells, thereby inducing immunosuppressive signals. These findings suggest potential therapeutic strategies targeting FXYD5-based mechanisms. Ru Wen, G Edward W. Marti, Neeladrisingha Das, Zhengyuan Qiu, Nathan Lam, Eric E. Peterson, Zenghua Fan, Aram Lyu, Fernando Jose Garcia Marques, Abel Bermudez, Hongjuan Zhao, Lawrence Fong, Guillem Pratx, Donna M. Peehl, Sharon J. Pitteri, James D. Brooks. FXYD5 plays diverse roles in immune evasion and tumor progression in prostate cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 4325.