In this meta-analysis, moderate interobserver agreement was found between expert thoracic radiologists interpreting high-resolution CT of interstitial lung disease, highlighting a need for greater standardization.

Neutrophil extracellular traps (NETs) play an essential role in antimicrobial defense. However, NETs have also been shown to promote and mediate a wide spectrum of diseases, including cancer, diabetes mellitus, cardiovascular diseases, and ocular diseases. Data regarding NETs in ocular diseases remain limited. In physiological conditions, NETs protect the eye from debris and cleave proinflammatory cytokines, including several interleukins. On the other hand, NETs play a role in corneal diseases, such as dry eye disease and ocular graft-versus-host disease, where they promote acinar atrophy and delayed wound healing. Additionally, NET levels positively correlate with increased severity of uveitis. NETs have also been described in the context of diabetic retinopathy. Although increased NET biomarkers are associated with an increased risk of the disease, NETs also assist in the elimination of pathological blood vessels and the regeneration of normal vessels. Targeting NET pathways for the treatment of ocular diseases has shown promising outcomes; however, more studies are still needed in this regard. In this article, we summarize the literature on the protective roles of NETs in the eye. Then, we describe their pathogenetic effects in ocular diseases, including those of the cornea, uvea, and retinal blood vessels. Finally, we describe the therapeutic implications of targeting NETs in such conditions.

Background: Generative models that synthesise virtual multiplex immunofluorescence (mIF) from haematoxylin and eosin (H&E) slides promise population-scale spatial proteomics. GigaTIME, trained on 40 million paired cells across 21 proteins in a US health system, is the largest such model. Whether its predictions reproduce independently measured protein outside its training distribution is untested against paired ground truth.Methods: We externally validated the released GigaTIME model, without retraining, across four cohorts absent from its training. Three carried paired protein ground truth: ORION (41 colorectal whole-slide images, high-plex mIF), HEMIT (15 colorectal regions, multiplex immunohistochemistry), and a head-and-neck squamous cell carcinoma (HNSCC) cohort given haematoxylin-only, out-of-distribution input. SurGen (423 colorectal cases) carried clinical and molecular labels only. We measured per-tile spatial fidelity, abundance concordance with and without nuclei (DAPI) partialled out, and pre-registered associations with mismatch-repair status and five-year survival.Findings: Only DAPI, the nuclear counterstain, was reproduced spatially (ORION r = 0·55; HEMIT r = 0·72). Every immune and functional marker fell in the pre-specified poor band, including the immunotherapy-decision markers CD8 (r = 0·13) and PD-L1 (r = 0·15). Apparent abundance concordance collapsed once cellularity was controlled (CD3 0·40 to 0·02; Ki-67 0·62 to 0·14), leaving only epithelium-coupled PD-L1 and PanCK. In SurGen, no predicted protein was associated with mismatch-repair or driver-mutation status (all q > 0·35), and the pre-registered protective CD8-survival association reversed (odds ratio 1·32 per SD, 95% CI 1·08 to 1·61). On out-of-distribution HNSCC input, even DAPI collapsed (r = 0·20).Interpretation: Across four external cohorts and three analyses, GigaTIME recovered nuclear morphology and tumour-epithelium content but not immune-marker spatial biology; its predicted immune markers carried no external clinical signal. The intact DAPI control localises the failure to generalisation, not implementation. Virtual mIF should not be relied upon for immune-marker readouts outside its training distribution without local, paired ground-truth validation.