Osaka City University

Large-scale radiographic datasets often include errors in labels such as body parts or projection, which can undermine automated image analysis. Therefore, we aimed to develop and externally validate two deep-learning models-one for categorising radiographs by body part, and another for identifying projection and rotation of chest radiographs-using large, diverse datasets. We retrospectively collected radiographs from multiple institutions and public repositories. For the first model (Xp-Bodypart-Checker), we included seven categories (Head, Neck, Chest, Incomplete Chest, Abdomen, Pelvis, Extremities). For the second model (CXp-Projection-Rotation-Checker), we classified chest radiographs by projection (anterior-posterior, posterior-anterior, lateral) and rotation (upright, inverted, left rotation, right rotation). Both models were trained, tuned, and internally tested on separate data, then externally tested on radiographs from different institutions. Model performance was assessed using overall accuracy (micro, macro, and weighted) as well as one-vs.-all area under the receiver operating characteristic curve (AUC). In the Xp-Bodypart-Checker development phase, we included 429,341 radiographs obtained from Institutions A, B, and MURA. In the CXp-Projection-Rotation-Checker development phase, we included 463,728 chest radiographs from CheXpert, PadChest, and Institution A. The Xp-Bodypart-Checker achieved AUC values of 1.00 (99% CI: 1.00-1.00) for all classes other than Incomplete Chest, which had an AUC value of 0.99 (99% CI: 0.98-1.00). The CXp-Projection-Rotation-Checker demonstrated AUC values of 1.00 (99% CI: 1.00-1.00) across all projection and rotation classifications. These models help automatically verify image labels in large radiographic databases, improving quality control across multiple institutions. Question This study examines how deep learning can accurately classify radiograph body parts and detect chest projection/orientation in large, multi-institutional datasets, enhancing metadata consistency for clinical and research workflows. Findings Xp-Bodypart-Checker classified radiographs into seven categories with AUC values of over 0.99 for all classes, while CXp-Projection-Rotation-Checker achieved AUC values of 1.00 across all projection and rotation classifications. Clinical relevance Trained on over 860,000 multi-institutional radiographs, our two deep-learning models classify radiograph body-part and chest radiograph projection/rotation, identifying mislabeled data and enhancing data integrity, thereby improving reliability for both clinical use and deep-learning research.

Infections are a major complication after allogeneic hematopoietic cell transplantation (allo-HCT). Although humoral immunity is essential, the phase-specific role of IgG remains unclear. We analyzed 579 allo-HCT recipients, stratified into early, intermediate, and late post-transplant phases. The incidence of bacterial infections during the intermediate phase decreased with increasing IgG trough levels (incidence rate ratio [IRR], 0.85; 95% confidence interval [CI]: 0.78-0.94, p < 0.01), whereas no association was observed in the early or late phases. During the intermediate phase, multivariable analyses demonstrated that hypogammaglobulinemia (IRR, 1.70; 95% CI: 1.11-2.62, p = 0.02) and acute graft-versus-host disease with grade II-IV (IRR, 2.10; 95% CI: 1.39-3.18, p < 0.01) were associated with bacterial infections. In models including high-dose steroid exposure, steroid therapy was independently associated with bacterial infections, whereas the effect of hypogammaglobulinemia was attenuated. There was a trend toward an interaction between hypogammaglobulinemia and high-dose steroid exposure. A stratified analysis suggested that hypogammaglobulinemia had a more pronounced impact among patients without high-dose steroids (IRR 1.86; 95% CI: 1.09-3.16; p = 0.02). These findings suggest the clinical relevance of hypogammaglobulinemia is phase-specific and influenced by the intensity of immunosuppressive therapy. Monitoring IgG during this period may help identify candidates for targeted immunoglobulin replacement therapy.

Mediastinoscopic esophagectomy for thoracic esophageal cancer may reduce pulmonary morbidity; however, upper mediastinal lymphadenectomy, particularly along the left recurrent laryngeal nerve (station 106recL), remains technically challenging. We developed a transhiatal-dominant mediastinoscopic subtotal esophagectomy using the da Vinci SP systemfor both the abdominal and cervical phases. After completing lower and mid-mediastinal lymphadenectomy while preserving the mediastinal pleura, transhiatal dissection is advanced beyond the carina into the upper mediastinum to perform 106recL lymphadenectomy along the left recurrent laryngeal nerve from its turning point beneath the aortic arch. The cervical phase is limited to confirming and completing the cranial 106recL dissection. Two patients had no postoperative morbidity, and bilateral vocal fold mobility was confirmed by laryngoscopy on postoperative day 3. Transhiatal/cervical console times were 180/71 and 200/45min, respectively,and postoperative hospital stay was 13days in both cases. This transhiatal-dominant, dual-phase da Vinci SP approach appears feasible for thoracic esophageal cancer.

Management of inflammatory bowel disease-associated gastrointestinal cancers, including Crohn's disease-associated cancer and ulcerative colitis-associated cancer, remains challenging because of concerns regarding treatment-related toxicity and limited clinical data. We aimed to investigate the safety of chemotherapy in patients with inflammatory bowel disease-associated gastrointestinal cancer. This retrospective study analyzed 21 patients with inflammatory bowel disease-associated gastrointestinal cancer (Crohn's disease-associated cancer: 10; ulcerative colitis-associated cancer: 11) who received chemotherapy or chemoradiotherapy between 2010 and 2024. Chemotherapy regimens, adverse events, treatment completion rates, and residual small intestinal lengths in patients with Crohn's disease were evaluated. Oxaliplatin-based chemotherapy was the most commonly used chemotherapeutic regimen. Dose reduction was required in all patients with Crohn's disease-associated cancer and in 64% of those with ulcerative colitis-associated cancer, primarily because of diarrhea or neurotoxicity. In the Crohn's disease-associated cancer group, diarrhea occurred in all patients and was significantly more severe in those with shorter residual small intestines (p=0.02). Despite these toxicities, 80% of patients with Crohn's disease-associated cancer and 55% of those with ulcerative colitis-associated cancer completed the planned chemotherapy regimen. No inflammatory bowel disease flares or need for corticosteroids or biologics occurred during treatment. Chemoradiotherapy was administered to five patients with Crohn's disease-associated cancer and locally advanced anorectal cancer, with no major safety concerns. Chemotherapy, including chemoradiotherapy, can be safely administered to patients with inflammatory bowel disease-associated gastrointestinal cancer with careful management. In Crohn's disease-associated cancer, residual small intestinal length is a key factor in predicting gastrointestinal toxicity severity, particularly diarrhea. Individualized dose adjustments and close monitoring are essential to balance cancer treatment efficacy with the unique risks associated with inflammatory bowel disease.

Abstract Loneliness in later life is a public‑health concern, and large language model (LLM)-enabled conversational robots have potential to supplement social connection at scale by facilitating everyday conversations. However, evidence on the development and acceptability of such systems for older people remains scarce. We conducted a two‑round mixed methods study with older community‑dwelling participants, to test whether pipeline refinements—streaming automatic speech recognition, text‑to‑speech, and simplified turn‑taking cues—improved conversational experience. A tabletop conversational robot (Sota, Vstone), powered by Generative Pre‑trained Transformer (GPT)‑4o was used. Eighteen individuals participated in round one; paired analyses were conducted on the 17 eligible participants remaining in round two. We collected interaction logs - (response latency, LLM‑level latency, and deletion rate (proportion of words in the manual transcript omitted in the ASR output), technical failure events, and participant questionnaires (Godspeed Questionnaire Series, GQS; System Usability Scale, SUS). Qualitative data were analysed using a framework approach. Median response latency decreased from 9.8 to 4.7s, deletion rate from 0.37 to 0.07, and usability (SUS) improved from 56.2 to 66.9. GQS subscales changed little and perceived companionship remained only modest. Qualitative accounts converged: interactions felt smoother and more in‑context, yet interruptions, perceived superficiality of conversation, and only a modest sense of companionship was still experienced. Findings indicate that improving basic conversational fluency enhances naturalness and usability but is insufficient for delivering truly relational experience outcomes. Progress toward better companionship requires an additional relational layer—shared memories, age‑attuned personas, and calibrated reciprocity.

Abstract Hippocampus is known to replay activity patterns to recall and process memories, which is often related to Hopfield-type attractor dynamics. Another line of theoretical studies suggests that hippocampal replay prioritizes replay of experiences to accelerate value learning for efficient decision making. It is unknown how hippocampal attractor dynamics perform prioritized memory sampling, and more broadly, how we can consistently relate dynamical (bottom-up) and functional (top-down) theories of hippocampal replay. In this paper, we propose an extended Hopfield-type attractor network model with momentum, kinetic energy, and conservation of the total energy, which is called momentum Hopfield model. We show that our model can be interpreted as CA3-CA1 network model with intrinsic oscillation, and such network model reproduces hippocampal replay in 1-D and 2-D spatial structures. We also prove that our model functionally works as Markov-chain Monte Carlo sampling in which recall frequencies of memory patterns can be arbitrarily biased. Using this property, we implemented prioritized experience replay using our model, which actually accelerated reinforcement learning for spatial navigation. Our model explains how dynamics of hippocampal circuits realize efficient memory sampling, providing a theoretical link between dynamics and functions of hippocampal replay.