Search

Bayesian deep learning is one of the key frameworks employed in handling predictive uncertainty. Variational inference (VI), an extensively used inference method, derives the predictive distributions by Monte Carlo (MC) sampling. The drawback of MC sampling is its extremely high computational cost compared to that of ordinary deep learning. In contrast, the moment propagation (MP)-based approach propagates the output moments of each layer to derive predictive distributions instead of MC sampling. Because of this computational property, it is expected to realize faster inference than MC-based approaches. However, the applicability of the MP-based method in deep models has not been explored sufficiently, even though some studies have demonstrated the effectiveness of MP only in small toy models. One of the reasons is that it is difficult to train deep models by MP because of the large variance in activations. To realize MP in deep models, some normalization layers are required but have not yet been studied. In addition, it is still difficult to design well-calibrated MP-based models, because the effectiveness of MP-based methods under various variational distributions has also not been investigated. In this study, we propose a fast and reliable MP-based Bayesian deep-learning method. First, to train deep-learning models using MP, we introduce a batch normalization layer extended to random variables to prevent increases in the variance of activations. Second, to identify the appropriate variational distribution in MP, we investigate the treatment of moments of several variational distributions and evaluate their uncertainty quality of predictions. Experiments with regression tasks demonstrate that the MP-based method provides qualitatively and quantitatively equivalent predictive performance to MC-based methods regardless of variational distributions. In the classification tasks, we show that we can train MP-based deep models by extended batch normalization. We also show that the MP-based approach realizes 2.0-2.8 times faster inference than the MC-based approach while maintaining the predictive performance. The results of this study can help realize a fast and well-calibrated uncertainty estimation method that can be deployed in a wider range of reliability-aware applications.

Background: JCOG1212 is a dose-finding and efficacy confirmatory study of concurrent superselective intra-arterial infusion of cisplatin and radiotherapy (RADPLAT) for locally advanced primary squamous cell carcinoma of the maxillary sinus (cT4a,bN0M0). In this study, we report the results of the final analysis of the efficacy confirmation phase for the T4a cohort with 5-year follow-up data to evaluate the late adverse events and long-term efficacy., Methods: Based on the results of the dose-finding phase, the efficacy confirmation phase consisted of seven weekly intra-arterial infusions of cisplatin 100 mg/m 2 combined with radiotherapy (70 Gy). The 5-year prognosis and late adverse events were evaluated., Results: Between April 2014 and August 2018, 64 patients were included in the analysis (one ineligible patient was excluded); 31 patients were treated with three-dimensional conformal radiation therapy (3D-CRT) and 33 with intensity modulated radiation therapy (IMRT). The 5-year overall survival, event-free survival, and local event-free survival was 71.9, 54.7, and 57.5%, respectively. In terms of late adverse events, grade 3 or higher non-hematologic toxicity was observed in 42.9% of 63 patients (retinopathy: 12, cataract: 10, osteonecrosis of mandible: 4, etc.). Grade 3 and 4 cataracts of affected side appeared in 22.6% (7/31) of the 3D-CRT group compared to 3.1% (1/32) in the IMRT group. Twenty-one patients had died, with 15 from the primary disease, 5 from other causes, and 1 from treatment-related cause., Conclusion: The prognosis of RADPLAT was favorable after 5-year follow-up with acceptable late adverse events and low proportion of treatment related death., Competing Interests: Declarations. Conflict of interest: A.H. reports grants and nonfinancial support from Japan Agency for Medical Research and Development (AMED) and the National Cancer Center Research and Development Fund during the conduct of the study. Y.S. reports personal fee from MSD.K.K. Ethics approval: This study was registered with the UMIN Clinical Trials Registry under trial number UMIN000013706 and the Japan Registry of Clinical Trials (Number: jRCTs031180004), and approved by the institutional review board from each participating institution and National Cancer Center Hospital Certified Review Board (CRB3180008). Consent to participate: Informed consent was obtained from all individual participants included in the study., (© 2025. The Author(s).)

In the 2019 Guidelines for the Management of Hypertension by the Japanese Society for Hypertension, lifestyle modification is recommended for all individuals except those with normal blood pressure. However, no detailed methods have been established to achieve the target blood pressure and resolve clinical inertia. CureApp HT, a digital therapeutic for hypertension that contributes to blood pressure reduction through lifestyle modification, was approved as software as a medical device for reimbursement by Japanese national health insurance in September 2022. This study aimed to survey physicians who implemented CureApp HT to assess how it changes physician-patient communication and contributes to clinical inertia resolution. A questionnaire survey was conducted at three time points: before the first prescription (first survey), 3 months (second survey), and 6 months (third survey) after the first prescription for physicians who had implemented CureApp HT. The primary outcome was the total score of five items on a Likert scale related to physician-patient communication, and it was analyzed based on the 47 physicians who responded to all three questionnaires. The total score of physician-patient communication significantly improved after 6 months of the introduction of CureApp HT, reflecting that physicians observed positive changes in patients' knowledge and attitudes regarding hypertension treatment. Furthermore, the number of physicians who set a target home blood pressure of 125/75 mmHg for their patients significantly increased. CureApp HT allows physicians to recognize changes in patients' disease knowledge and treatment attitudes, enabling them to set more stringent blood pressure targets and addressing clinical inertia. Physicians who implemented CureApp HT recognized changes in the patients' stages of behavioral change through improvements in patients' knowledge of the disease and their attitudes towards treatment, and by experiencing more effective communication, they set stricter blood pressure targets., Competing Interests: Compliance with ethical standards. Conflict of interest: Author TK received honoraria from CureApp, Inc. FH, MA, YH, YT, YI, and TT are employees of CureApp, Inc., (© 2024. The Author(s).)