This letter examines the interpretation and clinical translation of a meta-analysis of radiomics-based artificial intelligence (AI) for predicting pathological response after neoadjuvant immunochemotherapy in resectable non–small cell lung cancer. We highlight that the conventional-response comparator combines PERCIST and RECIST 1.1 assessments from the same 36-patient cohort, whereas the AI estimates arise from unmatched cohorts; consequently, the reported denominator and Z tests do not establish comparative superiority. We further consider how restriction to patients who reached resection and variation in imaging timepoints narrow the clinical estimand and limit inference about earlier treatment redirection. Clarification of the RECIST and error-direction examples is also warranted. We propose same-patient comparisons in treatment-initiation cohorts using fixed imaging times, locked thresholds, explicit primary-tumor and nodal labels, and calibration and net-benefit analyses at prespecified clinical thresholds.

arXiv:2603.07472v1 Announce Type: cross Abstract: In this study, we present a conditional diffusion-transformer framework for generating ensembles of three-dimensional Escherichia coli genome conformations guided by Hi-C contact maps. Instead of producing a single deterministic structure, we formulate genome reconstruction as a conditional generative modeling problem that samples heterogeneous conformations whose ensemble-averaged contacts are consistent with the input Hi-C data. A synthetic dataset is constructed using coarse-grained molecular dynamics simulations to generate chromatin ensembles and corresponding Hi-C maps under circular topology. Our models operate in a latent diffusion setting with a variational autoencoder that preserves per-bin alignment and supports replication-aware representations. Hi-C information is injected through a transformer-based encoder and cross-attention, enforcing a physically interpretable one-way constraint from Hi-C to structure. The model is trained using a flow-matching objective for stable optimization. On held-out ensembles, generated structures reproduce the input Hi-C distance-decay and structural correlation metrics while maintaining substantial conformational diversity, demonstrating the effectiveness of diffusion-based generative modeling for ensemble-level 3D genome reconstruction.