Facial asymmetry is relatively common in the general population. Here, we propose a fully automated annotation system that supports analysis of mandibular deviation and detection of facial asymmetry in posteroanterior (PA) cephalograms by means of a deep learning-based convolutional neural network (CNN) algorithm. In this retrospective study, 400 PA cephalograms were collected from the medical records of patients aged 4 years 2 months–80 years 3 months (mean age, 17 years 10 months; 255 female patients and 145 male patients). A deep CNN with two optimizers and a random forest algorithm were trained using 320 PA cephalograms; in these images, four PA landmarks were independently identified and manually annotated by two orthodontists. The CNN algorithms had a high coefficient of determination (R2), compared with the random forest algorithm (CNN-stochastic gradient descent, R2 = 0.715; CNN-Adam, R2 = 0.700; random forest, R2 = 0.486). Analysis of the best and worst performances of the algorithms for each landmark demonstrated that the right latero-orbital landmark was most difficult to detect accurately by using the CNN. Based on the annotated landmarks, reference lines were defined using an algorithm coded in Python. The CNN and random forest algorithms exhibited similar accuracy for the distance between the menton and vertical reference line. Our findings imply that the proposed deep CNN algorithm for detection of facial asymmetry may enable prompt assessment and reduce the effort involved in orthodontic diagnosis.

Aims Cardiohepatic interactions have been a focus of attention in heart failure (HF). The model for end‐stage liver disease excluding international normalized ratio (MELD‐XI) score has been shown to be useful for predicting poor outcomes in patients with acute decompensated HF (ADHF). Furthermore, the fibrosis‐4 (FIB‐4) index, a simple marker to assess liver fibrosis, predicts adverse prognoses in patients with HF as well. However, there is little information available on the prognostic significance of the combination of the MELD‐XI score and FIB‐4 index in patients with ADHF and its association with left ventricular ejection fraction (LVEF) subgroup. Methods and results We prospectively studied 466 consecutive patients who were admitted for ADHF [HF with reduced LVEF (LVEF < 40%): n = 164, HF with mid‐range LVEF (40% ≤ LVEF < 50%): n = 104, and HF with preserved LVEF (LVEF ≥ 50%): n = 198]. We calculated the MELD‐XI score and FIB‐4 indices at discharge. The primary endpoint was all‐cause death (ACD). During the mean follow‐up period of 2.8 years, 143 patients had ACD. In the multivariate Cox analysis, the MELD‐XI score and FIB‐4 index were independently associated with ACD. Patients were stratified into the following three groups according to the median value of MELD‐XI score (=11) and FIB‐4 index (=2.13): Group 1 had both a low MELD‐XI score and a low FIB‐4 index; Group 2 had either a high MELD‐XI score (MELD‐XI score ≥11) or a high FIB‐4 index (FIB‐4 index ≥2.13); and Group 3 had both a high MELD‐XI score and a high FIB‐4 index. Kaplan–Meier analysis revealed that Group 2 and Group 3 had a significantly greater risk of ACD than Group 1 [Group 2 vs. Group 1: adjusted hazard ratio, 2.48 (95% confidence interval: 1.75–3.53), P < 0.0001; Group 3 vs. Group 1: adjusted hazard ratio, 7.03 (95% confidence interval: 3.95–13.7), P < 0.0001]. In addition, the patients with both a higher MELD‐XI score and FIB‐4 index showed a significantly higher risk of ACD also in the patients with HF with reduced LVEF, HF with mid‐range LVEF, and HF with preserved LVEF (all P < 0.0001). Conclusions The combination of MELD‐XI score and FIB‐4 index may be useful for stratifying patients at risk for ACD in patients with ADHF, irrespective of LVEF.

Transfer learning (TL) is an alternative approach to the full training of deep learning (DL) models from scratch and can transfer knowledge gained from large-scale data to solve different problems. ImageNet, which is a publicly available large-scale dataset, is a commonly used dataset for TL-based image analysis; many studies have applied pre-trained models from ImageNet to clinical prediction tasks and have reported promising results. However, some have questioned the effectiveness of using ImageNet, which consists solely of natural images, for medical image analysis. The aim of this study was to evaluate whether pre-trained models using RadImageNet, which is a large-scale medical image dataset, could achieve superior performance in classification tasks in dental imaging modalities compared with ImageNet pre-trained models. To evaluate the classification performance of RadImageNet and ImageNet pre-trained models for TL, two dental imaging datasets were used. The tasks were (1) classifying the presence or absence of supernumerary teeth from a dataset of panoramic radiographs and (2) classifying sex from a dataset of lateral cephalometric radiographs. Performance was evaluated by comparing the area under the curve (AUC). On the panoramic radiograph dataset, the RadImageNet models gave average AUCs of 0.68 ± 0.15 (p < 0.01), and the ImageNet models had values of 0.74 ± 0.19. In contrast, on the lateral cephalometric dataset, the RadImageNet models demonstrated average AUCs of 0.76 ± 0.09, and the ImageNet models achieved values of 0.75 ± 0.17. The difference in performance between RadImageNet and ImageNet models in TL depends on the dental image dataset used.

We recently reported an unexpected role of osteoblast-derived matrix vesicles in the delivery of microRNAs to bone matrix. Of such microRNAs, we found that miR-125b inhibited osteoclast formation by targeting Prdm1 encoding a transcriptional repressor of anti-osteoclastogenesis factors. Transgenic (Tg) mice overexpressing miR-125b in osteoblasts by using human osteocalcin promoter grow normally but exhibit high trabecular bone mass. We have now further investigated the effects of osteoblast-mediated miR-125b overexpression on skeletal morphogenesis and remodeling during development, aging and in a situation of skeletal repair, i.e., fracture healing. There were no significant differences in the growth plate, primary spongiosa or lateral (periosteal) bone formation and mineral apposition rate between Tg and wild-type (WT) mice during early bone development. However, osteoclast number and medial (endosteal) bone resorption were less in Tg compared to WT mice, concomitant with increased trabecular bone mass. Tg mice were less susceptible to age-dependent changes in bone mass, phosphate/amide I ratio and mechanical strength. In a femoral fracture model, callus formation progressed similarly in Tg and WT mice, but callus resorption was delayed, reflecting the decreased osteoclast numbers associated with the Tg callus. These results indicate that the decreased osteoclastogenesis mediated by miR-125b overexpression in osteoblasts leads to increased bone mass and strength, while preserving bone formation and quality. They also suggest that, in spite of the fact that single miRNAs may target multiple genes, the miR-125b axis may be an attractive therapeutic target for bone loss in various age groups.