Conquering class imbalances in deep learning-based segmentation of dental radiographs with different loss functions.

OBJECTIVE: The imbalanced nature of real-world datasets is an ongoing challenge in the field of machine and deep learning. In medicine and in dentistry, most data samples represent patients not affected by pathologies, and on imagery, pathologic image areas are often smaller than healthy ones. Selecting suitable loss functions during deep learning is essential and may help to overcome the resulting imbalance. We assessed six different loss functions for one exemplary task, tooth structure segmentation on bitewing radiographs, for their performance. METHODS: Six different loss functions (Focal Loss, Dice Loss, Tversky Loss and hybrid losses of Cross-Entropy and Dice Loss, Focal and Dice Loss, Focal and Generalized Dice Loss) were compared on a tooth structure segmentation task of 1,625 bitewing radiographs. Training was performed using three different model architectures (U-Net, Linknet, DeepLavbV3+) over a 5-fold cross-validation. Tooth structures consisted of the classes (occurrence in% of samples/captures areas measured on pixel level) enamel (100 %/25 %), dentin (100 %/50 %), root canal (100 %/10 %), filling (81 %/8 %) and crown (28 %/5 %). RESULTS: Hybrid loss functions significantly outperformed standalone ones and provided robust results over the different architectures for the classes enamel, dentin, root canal and filling. Specifically, the Dice Focal loss reached high performance to conquer both image level and pixel level class imbalance, respectively. CLINICAL SIGNIFICANCE: In dental use cases it is often important to predict minority classes such as pathologies accurately. Using specific loss function may be an effective strategy to overcome data imbalance when training deep learning models.

Staphylococcus aureus small colony variants: A potentially underestimated microbiological challenge in peritoneal dialysis.

INTRODUCTION: Peritonitis remains the major infectious complication in the setting of peritoneal dialysis (PD). Despite known only moderate pathogenicity, the most frequently detected pathogens in PD-related peritonitis are surprisingly coagulase-negative staphylococci. However, this could be explained, at least in part, by Staphylococcus aureus small colony variants (SCVs) induced by PD fluids (PDFs) and misidentified by routinely used microbiological methods. MATERIAL AND METHODS: Bacteria were exposed to commonly used PDFs in various regimens designed to simulate daily use as closely as possible. Wild-type isolates and SCVs were subsequently used to determine minimum inhibitory concentrations (MICs), in vitro biofilm formation capacities, and auxotrophies. Underlying genetic alterations were investigated using whole-genome sequencing, and various microbial identification methods were tested to determine their performance for wild-types and SCVs. RESULTS: Stable SCVs could be isolated most successfully after exposure to glucose-containing PDFs alone. The reading of MICs was significantly affected by the reduced growth of SCVs, resulting in lower MIC values in 44% of all tests. Nonsynonymous mutations were found in all but one SCV, while only two isolates showed typical auxotrophic responses. While MALDI-TOF, PCR and Pastorex Staph-Plus correctly identified all S. aureus SCVs, API-Staph and VITEK-2 yielded identification rates of only 40% and 10%, respectively. CONCLUSIONS: Overall, the present study has shown that commercially available PDFs induce S. aureus SCVs in vitro, which are difficult to identify and test for antimicrobial susceptibility and can potentially lead to recurrent or persistent infections. Thus, they represent a potentially underappreciated challenge not only for microbiologists, but also for clinicians.

Rituximab treatment in pediatric-onset multiple sclerosis.

BACKGROUND AND PURPOSE: Rituximab (RTX) is frequently used off-label in multiple sclerosis. However, studies on the risk-benefit profile of RTX in pediatric-onset multiple sclerosis are scarce. METHODS: In this multicenter retrospective cohort study, patients with pediatric-onset multiple sclerosis from Sweden, Austria and Germany, who received RTX treatment were identified by chart review. Annualized relapse rates, Expanded Disability Status Scale scores and magnetic resonance imaging parameters (new T2 lesions and contrast-enhancing lesions) were assessed before and during RTX treatment. The proportion of patients who remained free from clinical and disease activity (NEDA-3) during RTX treatment was calculated. Side effects such as infusion-related reactions, infections and laboratory abnormalities were assessed. RESULTS: Sixty-one patients received RTX during a median (interquartile range) follow-up period of 20.9 (35.6) months. The annualized relapse rate decreased from 0.6 (95% confidence interval [CI] 0.38-0.92) to 0.03 (95% CI 0.02-0.14). The annual rate of new T2 lesions decreased from 1.25 (95% CI 0.70-2.48) to 0.08 (95% CI 0.03-0.25) and annual rates of new contrast-enhancing lesions decreased from 0.86 (95% CI 0.30-3.96) to 0. Overall, 70% of patients displayed no evidence of disease activity (NEDA-3). Adverse events were observed in 67% of patients. Six patients discontinued treatment due to ongoing disease activity or adverse events. CONCLUSION: Our study provides class IV evidence that RTX reduces clinical and radiological activity in pediatric-onset multiple sclerosis.

What is in a smile: The role of evaluation goal and response labels in facial muscle responses to prejudiced groups.

Based on the assumption that valence is permanently linked to facial responses, we expected that the corrugator muscle is contracted faster in response to overweight persons than to slim persons, whereas we expected faster contractions of the zygomaticus muscle in response to slim persons rather than to overweight persons. To detect such differences, we conducted experiments with different versions of a facial stimulus-response compatibility task that required participants to respond with the two facial muscles to photos of overweight or slim persons. Contrary to the assumption that valence is permanently linked to facial responses, in Experiments 1 and 2, social categories (overweight vs. slim persons) did not influence the response latencies assessed by electromyography. Whereas in Experiments 1 and 2, neutral labels were used for the muscle responses, in Experiment 3, affective response labels (smile vs. frown) were used. In Experiment 3, faster responses with the corrugator to overweight than to slim persons and with the zygomaticus to slim than to overweight persons were obtained. The influence of task and response label is consistent with the theory of event coding that suggests a more flexible link between valence and action.