A. Tsirigos

Spatially resolved tissue architecture and computational pathology in pancreatic cancer

Pancreatic ductal adenocarcinoma (PDAC) is a complex disease characterized by high levels of cellular heterogeneity and pronounced microenvironmental remodelling. Dynamic changes during its initiation and progression contribute to resistance to conventional therapies. Building upon key molecular catalogues established by bulk and single-cell profiling studies that have advanced our understanding of PDAC biology, recent advances in spatial biology have provided much-needed insights by elucidating regionally compartmentalized transcriptomic and proteomic programmes within the PDAC microenvironment. In parallel, emerging computational frameworks in digital pathology and artificial intelligence have advanced the field into a high-dimensional, quantitative discipline, particularly for classifying molecular and clinical features from histopathology images. Despite these advancements, integration of these two modalities remains a major challenge. Here, we summarize the convergence of molecular features identified through spatially resolved profiling in PDAC and its precursor lesions, as well as current developments in AI-powered pathology in cancer research. We further propose a multi-modal integration framework that maps molecular states onto morphological and architectural phenotypes, offering a roadmap for spatially informed patient stratification beyond descriptive tissue characterization. We posit that the path forward relies on disciplined cross-scale integration of spatial, histological, and clinical data to ensure meaningful translation into clinical practice. Pancreatic ductal adenocarcinoma (PDAC) is a highly lethal cancer, often diagnosed late and resistant to treatment. This review highlights emerging evidence from spatial profiling studies demonstrating that PDAC and its precursor lesions are not just a simple mixture of malignant and stromal cells but rather a structured ecosystem with spatially distinct immune and fibroblast niches. Digital pathology and artificial intelligence offer promising complementary approaches to extending insights gained from spatial omics to larger patient cohorts, with the potential to improve risk stratification, prognostic prediction and assessment of therapeutic response. Future research should focus on establishing ground-truth characteristics in different disease contexts in the pancreas and expanding spatially resolved datasets across diverse PDAC cohorts to enable robust development of computational frameworks and improve their clinical applications.

Genome Profiling of Actionable Cancer Targets (NYU LG-PACT) for Clinical Patient Molecular Diagnostics and Treatment

Vision-Language Models for Image-Based Dietary Assessment: A Benchmark of Accuracy, Cost, and Prompt Strategies Across Ten Models

Background Dietary assessment is the cornerstone of clinical management and research studies evaluating diet and health. Traditional methods such as food diaries and 24-hour recalls can be burdensome, prone to recall bias, and difficult to adhere to. Image-based dietary assessment using vision-language models (VLMs) offers a potential solution. Objective Our goal was to benchmark state-of-the-art VLMs for automated food recognition, weight estimation, and calorie estimation using Google’s Nutrition5k dataset. Methods We evaluated 3,229 food images using ten approaches: proprietary VLMs (Gemini 2.0 Flash, 2.5 Flash, 3.0 Flash, and 3.1 Flash Lite; GPT- 4o, GPT-4o-mini, and GPT-5 Mini; and Claude Haiku 4.5), an open-source VLM (Qwen2-VL-7B), and a commercial food recognition API (FatSecret). We assessed calorie and weight estimation using Lin’s Concordance Correlation Coefficient (CCC) and component detection using Jaccard similarity. Results Gemini 3.0 Flash achieved the best calorie estimation (CCC 0.767, MAE 80.7 kcal), while Gemini 3.1 Flash Lite offered very comparable accuracy (CCC 0.754) with the highest ingredient recognition (Jaccard 0.655) at the lowest cost among top-performing models ($0.59/1K images). Among earlier-generation models, Gemini 2.0 Flash remained competitive (CCC 0.742, Jaccard 0.621) at a fraction of the cost ($0.10/1K images). A human validation study in which four annotators reviewed 440 images revealed systematic omissions in the original Nutrition5k labels. After correction, the extrapolated ingredient-overlap score for Gemini 2.0 Flash increased from 0.62 to an estimated 0.82, suggesting that raw Jaccard scores substantially underestimate true model performance. Conclusions Current VLMs can perform automated dietary assessment with reasonable accuracy from single overhead photographs. Our results inform model selection for dietary assessment applications and highlight remaining challenges in calorie estimation and component detection for complex, multi-item meals.