University of New Haven

Congenital lung abnormalitiesare commonly encountered in the recent era of multi-modality fetal imaging. Congenital lung abnormalities encompass a diverse range of developmental disorders affecting the lung parenchyma, airways, and vascular structures. The anomalies, such as congenital pulmonary airway malformations (CPAM), bronchopulmonary sequestration, and congenital lobar overinflation (CLO), can present with varying degrees of severity, from asymptomatic cases to life-threatening respiratory distress. Ultrasound (US) plays a crucial role in the prenatal detection of these abnormalities, providing real-time imaging and facilitating early diagnosis and management planning. Magnetic resonance imaging (MRI), on the other hand, provides detailed morphological insights, particularly for complex cases. Its ability to visualize soft tissue structures and vascular anomalies without ionizing radiation makes it an indispensable tool in fetal and pediatric imaging. The combined use of US and MRI enhances diagnostic accuracy for fetuses with congenital lung abnormalities. These modalities continue to evolve, offering promising advancements in the understanding and management of these conditions. In this review article, we present a spectrum of common and uncommon congenital lung abnormalities, along with their typical imaging findings in US and MRI, and an approach to diagnosis.

Abstract Introduction ccRCC exhibits diverse tumor microenvironments, and associations with ICI response have been demonstrated using bulk RNA sequencing and molecular subtype analyses. However, the factors that drive this microenvironmental heterogeneity remain unclear. This study aimed to analyze large-scale single-cell data to identify multicellular co-occurrence patterns within the tumor microenvironment and the tumor-intrinsic programs that shape these ecosystems. Methods We analyzed single-cell RNA-seq data from 58 ccRCC cases (224,018 cells), including 6 untreated, 31 pre-ICI, 14 post-ICI, 3 pre-TKI, and 4 post-TKI (Kashima, ASCO 2024). We performed high-resolution clustering for each non-tumor cell lineage (CD8+ T, CD4+ T, regulatory T cells, myeloid cells, endothelial cells, and fibroblasts). Pairwise correlations of cluster frequencies across samples were computed to detect “modules” representing co-existing immune and stromal populations. These modules were validated using an independent single-cell RNA-seq dataset (HCRN GU16-260, N = 16; Hugaboom, Cancer Discov. 2025). Tumor cells (50,346 cells) were separately analyzed using consensus NMF to identify shared gene programs. We compared the prevalence of modules in responders (R) (n = 10; complete or partial response) with non-responders (NR) (n = 8; progressive disease) to ICI therapies. Results We identified 78 fine clusters and 21 co-existence modules. These clusters and modules captured features not fully explained by bulk RNA-seq or small-scale single-cell data. Most modules consisted of multiple cell types rather than a single lineage, i.e., a CXCL13+ CD4+ T cell–memory B cell module. Module structure was supported in the HCRN cohort (Mantel r = 0.3, P < 0.001). Patients exhibited highly heterogeneous combinations of these modules. Several modules were associated with favorable ICI responses, including exhausted CD8+ T cells–MACRO+ macrophages (Module R1) and cDC–iCAF (Module R2). Conversely, modules such as Th17 CD4+ T cells–GNLY+ ILCs–pDCs (Module NR1) and GZMA+ CD4+ T cells–Trm CD8+ T cells–IFNγ+ ILCs (Module NR2) were associated with poor ICI response. Consensus NMF of tumor cells revealed six interpretable gene programs shared across samples. Notably, the tumor-intrinsic IL6 and VEGF programs were correlated with Module NR1, while Module NR2 was mainly associated with the MYC program. Modules R1 and R2 also demonstrated distinct associations with specific tumor gene programs. These findings suggest that patient-specific, multiple biological pathways may contribute to both resistance and responsiveness to ICI therapy. Conclusion Our large-scale single-cell analysis identifies reproducible multicellular modules within the ccRCC tumor microenvironment, several of which are positively or negatively associated with ICI response. The observed relationships between non-tumor cell modules and tumor-intrinsic gene programs indicate that tumor cell states and microenvironmental composition are interdependent, highlighting potential cellular crosstalk that may influence immunotherapy efficacy. Citation Format: Katsuhiro Ito, Soki Kashima, Rishabh Rout, Zhaochen Ye, Nicholas R. Schindler, David A. Braun. Multicellular modules and tumor programs shape the tumor microenvironment and predict immune checkpoint inhibitor (ICI) response in clear cell renal cell carcinoma (ccRCC) [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Innovations in Kidney Cancer Research: From Molecular Insights to Therapeutic Breakthroughs; 2026 Mar 13-16; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2026;86(5_Suppl_2):Abstract nr A009.

Artificial intelligence has transformed phishing from opportunistic deception into automated, adaptive social engineering on a scale. Contemporary campaigns leverage generative content synthesis, adversary-in-the-middle credential relay, QR-code mobile pivots, and infrastructure churn to evade traditional signature-based and reputation-driven defenses. As identity systems increasingly underpin cloud services and critical infrastructure, phishing mitigation becomes a resilience challenge rather than a purely technical filtering problem. This study introduces the AID-PDR Framework (AI-Driven Phishing Defense & Resilience), a multi-layer socio-technical architecture integrating phishing-resistant authentication (FIDO2/WebAuthn), standards-based email authentication (SPF, DKIM, DMARC with MTA-STS and TLS-RPT), browser-level enforcement, adversarially robust machine learning detection pipelines, and adaptive AI-driven phishing simulation. A quantitative resilience model illustrates how layered controls produce multiplicative risk reduction across independent defensive mechanisms. Grounded in breach investigations, threat intelligence reporting, applied ML design patterns, and alignment with NIST and Zero Trust frameworks, this work advances a unified approach to mitigating AI-driven phishing risk at enterprise and national scale.

The growing number of older bridges has resulted in an increase in structural flaws, demanding frequent inspections and maintenance. Structural degradation accelerates post-damage recovery, emphasizing the necessity of preventive interventions. The use of Uncrewed Aerial Vehicle Systems (UASs) for bridge inspections represents a significant development in structural health monitoring (SHM). Traditional inspection methods are labor-intensive, time-consuming, expensive, and require access to high or difficult-to-reach areas, posing safety risks to inspectors. This study focuses on identifying drones that can efficiently support bridge inspection activities. Key factors influencing UAS selection include flight performance, flying modes, cost, sensor capabilities, payload capacity, and controller communication. The primary objective of this paper is to provide guidance to inspectors and transportation agencies regarding the capabilities and limitations of commercially available drones. It also outlines potential cost considerations associated with drone selection, including pilot skill level, platform cost, and sensor integration. These factors may vary depending on the type and complexity of the bridge being inspected. By addressing these aspects, this paper aims to assist decision-makers in making informed choices regarding the use of UASs for bridge inspection applications.

Abstract Introduction: The I-SPY2 neoadjuvant platform trial (NCT01042379) evaluates new therapies in high molecular risk, stage 2-3 breast cancer patients. Between 2010 and 2022, 24 therapies were tested in 2,117 patients. One-third of these patients had HER2-negative tumors with low expression of immune genes and no DNA repair deficiency (DRD) (HER2-/Immune-/DRD- subtype). These patients had very low pathologic complete response (pCR) rates (∼10% on average) across all treatment and control arms. We performed a high-throughput drug screen (n=386 compounds) in triple-negative (TN), immune-low, DRD-low breast cancer patient-derived organoids with confirmed in vitro resistance to platinum chemotherapy to identify effective treatments for this tumor type. HSP90 inhibitors, including ganetespib, were one of the most potent drug classes in the screen. HSP90 is an abundant chaperone protein involved in the folding of oncoproteins under basal conditions and during stress response. Neoadjuvant ganetespib was tested in I-SPY2 in combination with standard chemotherapy in HER2-negative breast cancer but did not meaningfully improve pCR rates (PMID 36456573). Here, we re-evaluate the ganetespib arm by Response Predictive Subtypes with 5-year median follow-up data, focused on patients with breast tumors of the HER2-/Immune-/DRD- subtype. Methods: HER2-/Immune-/DRD- patients (N=221; 79% HR+, 21% TN) and HER2-/Immune+ and/or DRD+ patients (N=219; 32% HR+, 68% TN) were randomized to neoadjuvant ganetespib (150 mg/m2 ganetespib every 3 weeks with paclitaxel, then doxorubicin plus cyclophosphamide (AC)) (n=38 HER2-/Immune-/DRD- and n=55 HER2-/Immune+ and/or DRD+), or standard neoadjuvant chemotherapy control (paclitaxel-AC) (n=183 HER2-/Immune-/DRD- and n=164 HER2-/Immune+ and/or DRD+). The primary endpoint, pCR, was defined as the absence of invasive disease in the breast and regional nodes at surgery. Distant relapse-free survival (DRFS) was assessed using Kaplan-Meier curves, with hazard ratios (HR) estimated by Cox regression. Residual cancer burden (RCB) and serial MRI-based functional tumor volume (FTV) were collected as investigational early measures of treatment efficacy in the neoadjuvant setting. Results: In HER2-/Immune-/DRD- patients, pCR rates were low in the ganetespib and control arms (13% vs. 9%), compared with 38% vs. 31%, respectively, in HER2-/Immune+ and/or DRD+ patients. In both subtypes, patients who achieved a pCR with either treatment had excellent DRFS (100% at 5 years). Interestingly, in HER2-/Immune-/DRD- patients treated with ganetespib with no pCR (n=32), we observed only 1 DRFS event, compared with 35 DRFS events in HER2-/Immune-/DRD- control-treated patients with no pCR (n=166); HR 0.12 (95% CI 0.02-0.90; p-value: 0.039). The respective DRFS at 5 years was 96% with ganetespib vs. 77% with control, showing a significantly improved DRFS with neoadjuvant ganetespib for HER2-/Immune-/DRD- patients, despite their non-pCR. This treatment effect remained significant when adjusted for hormone receptor status. Among non-pCR patients, RCB measure and FTV decrease explained a portion of ganetespib’s significant DRFS benefit. In contrast, ganetespib did not improve DRFS in HER2-/Immune+ and/or DRD+ patients compared with control. Conclusion: Our data suggest a promising role for ganetespib in improving long-term outcomes for patients with HER2-/Immune-/DRD- tumors, which account for 38% of all molecularly high-risk breast cancers, even in the absence of a pathologic complete response. Continuous measures of RCB and FTV may serve as early predictors of ganetespib’s long-term survival benefit in HER2-/Immune/-DRD- patients. Citation Format: T. B. V. Bui, D. M. Wolf, C. Yau, R. W. Sayaman, K. Chow, S. W. Choi, M. C. Bruck, K. Santos-Parker, A. M. Glas, J. E. Lang, A. Forero-Torres, L. Brown-Swigart, G. L. Hirst, E. F. Petricoin, J. Perlmutter, W. F. Symmans, P. R. Pohlmann, D. Yee, L. Pusztai, I-SPY investigators, N. M. Hylton, A. M. DeMichele, J. M. Rosenbluth, L. J. Esserman, L. J. Van 'T Veer. Neoadjuvant HSP90 inhibition with ganetespib significantly improves distant relapse-free survival in I-SPY2 patients with HER2-negative breast cancer with low expression of immune genes and no DNA repair deficiency [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS3-11-10.

The risk of developing some types of head and neck squamous cell carcinoma (HNSCC) is seven times higher in males, and such disparities may not be associated only with tobacco and alcohol consumption or HPV infection. Therefore, the endocrine microenvironment is considered another risk factor, as epidemiologic studies have unequivocally shown the protective effect of estrogen in women. This research was focused on progesterone receptors (PRs), the least-studied sex hormone receptors in HNSCC. Our study included fresh tissue samples from 95 primary tumors, 25 metastatic lymph nodes and 40 healthy oral mucosa. Gene expression of nuclear (PGR) and seven membrane PRs (PAQR5, PAQR6, PAQR7, PAQR8, PAQR9, PGRMC1 and PGRMC2) was analyzed by qRT-PCR and associated with clinicopathological characteristics. The results showed that, compared to control tissue, PGR was increased in metastatic lymph nodes, while PAQR5, PAQR7, PAQR8 and PAQR9 were decreased in primary tumors (all p < 0.05). The expression of almost all PRs was greater in older patients and showed moderate to strong positive mutual correlations in both tumors and controls. PARQ8 and PAQR9 were increased in females and pT4 tumors (all p < 0.05). Survival analysis showed that higher PAQR5 (hazard ratio (HR) 2.8, 95% confidence interval (CI) 1.19-6.57, p = 0.019) and PAQR7 (HR 2.0, 95% CI 1.01-3.81, p = 0.048) were associated with worse overall survival, but their independence was not proven in multivariate analysis. Although most PRs were reduced in primary tumors, an increased PAQR5 expression, also associated with tumor invasion markers, could likely mark a specific aggressive, advanced stage of primary tumors and potentially serve as a negative prognostic biomarker for HNSCC.

The COVID-19 pandemic disrupted the global construction industry, resulting in workforce shortages, project delays, and contractual disputes. This study used an online survey of private engineering and construction companies to assess their responses to the challenges posed by COVID-19. The survey focused on workforce management, technological advancements, safety protocols, and project execution. Workforce adaptations, particularly the shift to telework, saw 66% of respondents reporting partial telework and 22% indicating that telework became permanent after the pandemic. Technological advancements were crucial for maintaining operations during the pandemic, with 82% of respondents reporting the adaptation of new digital tools. These included widespread adoption of Microsoft Teams and Zoom by both office and field staff. Training was identified as essential for the effective implementation of these technologies, enabling employees to maximize their benefits. The pandemic affected project design and construction activities, with 69% of respondents reporting delays during the design phase and 70% during construction. Cost overruns of up to 25% were reported in the public sector and 67% of respondents in the private sector reported cost overruns, highlighting the financial impact of the pandemic. The findings provide insights into organizational resilience strategies adopted by private firms and offer lessons that can inform preparedness for future disruptions.

Deepfake-enabled identity abuse has moved from a peripheral cyber-risk to an operational threat for U.S. banking, especially in remote onboarding, account recovery, contact-center authentication, and high-risk payment authorization. The problem is no longer limited to obvious synthetic media. Financial institutions increasingly face blended attacks that combine forged identity documents, face or voice cloning, injected video streams, social-engineering pressure, mule accounts, and adaptive retries across channels. This paper develops a publication-ready research framework for machine learning systems that identify deepfake-driven customer impersonation, authentication evasion, and identity abuse in U.S. banking environments. The study synthesizes regulatory guidance, public fraud data, biometric spoofing research, deepfake detection literature, and banking fraud analytics to propose a multimodal detection architecture spanning document forensics, face presentation attack detection, deepfake-video detection, speaker anti-spoofing, behavioral biometrics, device and network telemetry, graph-based entity resolution, and risk-calibrated decisioning. Real public evidence motivates the design: FTC data show reported fraud losses reached $12.5 billion in 2024, including $2.95 billion in imposter-scam losses, while FinCEN reported increasing suspicious activity narratives involving deepfake media targeting financial institutions. The proposed methodology treats identity abuse as a sequential, multimodal, and adversarial classification problem rather than a single-screen biometric check. The paper argues that the most effective defense is not one detector but an explainable ensemble with smart friction, human escalation, and governance aligned to NIST identity-proofing standards, anti-money-laundering expectations, and consumer-protection obligations. By linking technical detection with operational controls, the study provides a practical blueprint for authentication, resilient fraud prevention, and more trustworthy digital banking systems.

<title>Abstract</title> Large language models are increasingly deployed in medical settings, yet their sensitivity to prompt formatting remains poorly characterized. We evaluate MedGemma (4B and 27B variants) on MedMCQA (4,183 questions) and PubMedQA (1,000 questions). Our experiments reveal concerning findings: chain- of-thought prompting decreases accuracy by 5.7% compared to direct answering; few-shot examples degrade performance by 11.9% while increasing position bias from 0.14 to 0.47; shuffling answer options causes the model to change predictions 59.1% of the time with accuracy dropping up to 27.4 percentage points; and truncating context to 50% causes accuracy to plummet below the no-context baseline. These results demonstrate that prompt engineering techniques validated on general-purpose models do not transfer to domain-specific medical LLMs.