Deep learning for accurately recognizing common causes of shoulder pain on radiographs.

OBJECTIVE: Training a convolutional neural network (CNN) to detect the most common causes of shoulder pain on plain radiographs and to assess its potential value in serving as an assistive device to physicians. MATERIALS AND METHODS: We used a CNN of the ResNet-50 architecture which was trained on 2700 shoulder radiographs from clinical practice of multiple institutions. All radiographs were reviewed and labeled for six findings: proximal humeral fractures, joint dislocation, periarticular calcification, osteoarthritis, osteosynthesis, and joint endoprosthesis. The trained model was then evaluated on a separate test dataset, which was previously annotated by three independent expert radiologists. Both the training and the test datasets included radiographs of highly variable image quality to reflect the clinical situation and to foster robustness of the CNN. Performance of the model was evaluated using receiver operating characteristic (ROC) curves, the thereof derived AUC as well as sensitivity and specificity. RESULTS: The developed CNN demonstrated a high accuracy with an area under the curve (AUC) of 0.871 for detecting fractures, 0.896 for joint dislocation, 0.945 for osteoarthritis, and 0.800 for periarticular calcifications. It also detected osteosynthesis and endoprosthesis with near perfect accuracy (AUC 0.998 and 1.0, respectively). Sensitivity and specificity were 0.75 and 0.86 for fractures, 0.95 and 0.65 for joint dislocation, 0.90 and 0.86 for osteoarthrosis, and 0.60 and 0.89 for calcification. CONCLUSION: CNNs have the potential to serve as an assistive device by providing clinicians a means to prioritize worklists or providing additional safety in situations of increased workload.

Deep-Learning-Based Diagnosis of Bedside Chest X-ray in Intensive Care and Emergency Medicine.

OBJECTIVES: Validation of deep learning models should separately consider bedside chest radiographs (CXRs) as they are the most challenging to interpret, while at the same time the resulting diagnoses are important for managing critically ill patients. Therefore, we aimed to develop and evaluate deep learning models for the identification of clinically relevant abnormalities in bedside CXRs, using reference standards established by computed tomography (CT) and multiple radiologists. MATERIALS AND METHODS: In this retrospective study, a dataset consisting of 18,361 bedside CXRs of patients treated at a level 1 medical center between January 2009 and March 2019 was used. All included CXRs occurred within 24 hours before or after a chest CT. A deep learning algorithm was developed to identify 8 findings on bedside CXRs (cardiac congestion, pleural effusion, air-space opacification, pneumothorax, central venous catheter, thoracic drain, gastric tube, and tracheal tube/cannula). For the training dataset, 17,275 combined labels were extracted from the CXR and CT reports by a deep learning natural language processing (NLP) tool. In case of a disagreement between CXR and CT, human-in-the-loop annotations were used. The test dataset consisted of 583 images, evaluated by 4 radiologists. Performance was assessed by area under the receiver operating characteristic curve analysis, sensitivity, specificity, and positive predictive value. RESULTS: Areas under the receiver operating characteristic curve for cardiac congestion, pleural effusion, air-space opacification, pneumothorax, central venous catheter, thoracic drain, gastric tube, and tracheal tube/cannula were 0.90 (95% confidence interval [CI], 0.87-0.93; 3 radiologists on the receiver operating characteristic [ROC] curve), 0.95 (95% CI, 0.93-0.96; 3 radiologists on the ROC curve), 0.85 (95% CI, 0.82-0.89; 1 radiologist on the ROC curve), 0.92 (95% CI, 0.89-0.95; 1 radiologist on the ROC curve), 0.99 (95% CI, 0.98-0.99), 0.99 (95% CI, 0.98-0.99), 0.98 (95% CI, 0.97-0.99), and 0.99 (95% CI, 0.98-1.00), respectively. CONCLUSIONS: A deep learning model used specifically for bedside CXRs showed similar performance to expert radiologists. It could therefore be used to detect clinically relevant findings during after-hours and help emergency and intensive care physicians to focus on patient care.

Highly accurate classification of chest radiographic reports using a deep learning natural language model pre-trained on 3.8 million text reports.

MOTIVATION: The development of deep, bidirectional transformers such as Bidirectional Encoder Representations from Transformers (BERT) led to an outperformance of several Natural Language Processing (NLP) benchmarks. Especially in radiology, large amounts of free-text data are generated in daily clinical workflow. These report texts could be of particular use for the generation of labels in machine learning, especially for image classification. However, as report texts are mostly unstructured, advanced NLP methods are needed to enable accurate text classification. While neural networks can be used for this purpose, they must first be trained on large amounts of manually labelled data to achieve good results. In contrast, BERT models can be pre-trained on unlabelled data and then only require fine tuning on a small amount of manually labelled data to achieve even better results. RESULTS: Using BERT to identify the most important findings in intensive care chest radiograph reports, we achieve areas under the receiver operation characteristics curve of 0.98 for congestion, 0.97 for effusion, 0.97 for consolidation and 0.99 for pneumothorax, surpassing the accuracy of previous approaches with comparatively little annotation effort. Our approach could therefore help to improve information extraction from free-text medical reports. Availability and implementationWe make the source code for fine-tuning the BERT-models freely available at https://github.com/fast-raidiology/bert-for-radiology. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Development of a Deep Learning Algorithm for Periapical Disease Detection in Dental Radiographs.

Periapical radiolucencies, which can be detected on panoramic radiographs, are one of the most common radiographic findings in dentistry and have a differential diagnosis including infections, granuloma, cysts and tumors. In this study, we seek to investigate the ability with which 24 oral and maxillofacial (OMF) surgeons assess the presence of periapical lucencies on panoramic radiographs, and we compare these findings to the performance of a predictive deep learning algorithm that we have developed using a curated data set of 2902 de-identified panoramic radiographs. The mean diagnostic positive predictive value (PPV) of OMF surgeons based on their assessment of panoramic radiographic images was 0.69(± 0.13), indicating that dentists on average falsely diagnose 31% of cases as radiolucencies. However, the mean diagnostic true positive rate (TPR) was 0.51(± 0.14), indicating that on average 49% of all radiolucencies were missed. We demonstrate that the deep learning algorithm achieves a better performance than 14 of 24 OMF surgeons within the cohort, exhibiting an average precision of 0.60(± 0.04), and an F1 score of 0.58(± 0.04) corresponding to a PPV of 0.67(± 0.05) and TPR of 0.51(± 0.05). The algorithm, trained on limited data and evaluated on clinically validated ground truth, has potential to assist OMF surgeons in detecting periapical lucencies on panoramic radiographs.

Chirurgie bei Läsionen des Nervus facialis.

There are numerous reasons for facial palsy, which range from idiopathic palsy (Bell's palsy) to destruction of the facial nerve by a malignant salivary gland tumor. If the chance of spontaneous recovery is low or there is no drug therapy available, surgery is a therapeutical option. Recently, larger studies were published by specialized centers which enable a more individualized therapeutical concept to achieve tone, symmetry and movement of the paralyzed face based on a detailed preoperative assessment. An important therapy target is the improvement of patient´s quality of life. In the present article, we systematically review the important diagnostic steps and, directly derived from this, the indications for surgical options for reanimation of the mimic function. Furthermore, we provide an overview about a variety of postoperative adjuvant measures as well as on new objective assessment tools to evaluate the therapy results.

Reciprocal Predictive Accuracy of Sinonasal Symptom Severity, Nasal Endoscopy, and Frequency of Past Chronic Rhinosinusitis Exacerbations.

Objective We sought to determine whether chronic rhinosinusitis (CRS) symptom severity, endoscopic exam findings, and acute exacerbation of CRS (AECRS) frequency-all important and distinct clinical manifestations of CRS-would be predictive of each other and, therefore, inform when further assessment of each other metric should be pursued. Study Design Cross-sectional cohort study. Setting Tertiary academic rhinology clinic. Subjects and Methods In total, 241 patients with CRS were prospectively recruited and completed the 22-item Sinonasal Outcome Test (SNOT-22) to reflect CRS symptom severity. AECRS frequency was assessed using the number of sinus infections as well as CRS-related antibiotics and CRS-related oral corticosteroids used in the past 3 months. An endoscopy score was calculated for each patient. Results SNOT-22 score and AECRS were predictive of each other while AECRS and endoscopy score were not predictive of each other. SNOT-22 score could be used to predict having had, in the past 3 months, at least 1 sinus infection (area under the curve [AUC] = 0.727; P < .001), at least 1 CRS-related antibiotic used (AUC = 0.691; P < .001), or at least 1 CRS-related oral corticosteroid course used (AUC = 0.655; P < .001). Having a SNOT-22 score ≥30 could be predicted by reporting at least 1 sinus infection (AUC = 0.634; P < .001), CRS-related antibiotics (AUC = 0.614; P < .001), or CRS-related oral corticosteroids (AUC = 0.616; P < .001) in the past 3 months. These relationships held for patients with and without nasal polyps. Conclusion The predictive power of CRS outcome measures reflecting symptomatology, AECRS frequency, and endoscopic findings may be of clinical utility in situations where time or resources are limited to perform an ideally full assessment of patients with CRS.

Impact of odontogenic chronic rhinosinusitis on general health-related quality of life.

Chronic rhinosinusitis (CRS) may arise due to odontogenic etiologies. However, it is unknown whether odontogenic CRS has a differential impact on patients' quality of life (QOL) compared to standard, inflammatory (but non-odontogenic) CRS. The objective of this study was to determine whether there is a difference in the impact of sinonasal symptomatology on general health-related QOL in odontogenic CRS compared to non-odontogenic CRS. This was a retrospective review of 21 odontogenic CRS patients who visited our tertiary care center. The severity of sinonasal symptomatology and CRS-specific QOL detriment was measured using the 22-item Sinonasal Outcomes Test (SNOT-22) and general health-related QOL was measured using the health utility index from the 5-item EuroQol survey (EQ-5D HUV). Compared to non-odontogenic CRS, odontogenic CRS was not associated with a difference in SNOT-22 score [linear regression coefficient (ß) = - 1.57, 95% CI - 12.47 to 9.32, p = 0.777] but was significantly associated with decreased EQ-5D HUV (ß = - 0.10, 95% CI - 0.17 to - 0.03, p = 0.008). We also found that the magnitude of association (ß) between SNOT-22 and EQ5D-HUV was greater for odontogenic CRS patients compared to non-odontogenic CRS patients (p = 0.045). Our findings suggest sinonasal symptoms may have a greater impact on general QOL in odontogenic CRS compared to non-odontogenic CRS. The reason for this remains unknown, but deserves further study.

Association between Asthma and Chronic Rhinosinusitis Severity in the Context of Asthma Control.

Objective Comorbid asthma is associated with decreased quality of life (QOL) in chronic rhinosinusitis (CRS). It is unclear whether this association is independent of the patients' clinical asthma status. We therefore sought to determine if asthma is associated with lower QOL in CRS, independent of asthma control. Study Design Cross-sectional cohort study of 350 patients with CRS. Setting Tertiary academic rhinology clinic. Subjects and Methods In total, 350 participants with CRS were recruited and 28.3% were asthmatic. CRS-specific QOL was measured using the 22-item Sinonasal Outcome Test (SNOT-22). Asthma control was assessed with the Asthma Control Test (ACT). General health-related QOL was assessed with the EuroQoL 5-dimensional general health-related quality of life survey visual analog scale (EQ-5D VAS). Associations were sought between SNOT-22 and EQ-5D VAS (dependent variables) and asthma (independent variable), while controlling for ACT. ACT score for patients with CRS without asthma was set at 25 (indicating completely controlled, asymptomatic asthma). Results Comorbid asthma was associated with SNOT-22 (ß = 11.8; 95% confidence interval [CI], 6.2-17.3; P < .001) and EQ-5D VAS (ß = -6.2; 95% CI, -11.2 to -1.3; P = .014). After controlling for ACT, asthma was no longer associated with SNOT-22 ( P = .147) or EQ-5D VAS ( P = .994). Instead, ACT score was associated with SNOT-22 (ß = -2.1; 95% CI, -3.2 to -1.1; P < .001) and EQ-5D VAS (ß = 2.1; 95% CI, 1.1 to 3.0; P < .001). ACT score completely drove the association between asthma and worse QOL. Conclusion Comorbid asthma is not necessarily reflective of decreased QOL in CRS. The association of comorbid asthma with lower QOL in CRS is related to the clinical status (eg, control) of asthma.

Association between symptoms of allergic rhinitis with decreased general health-related quality of life.

BACKGROUND: The impact of specific allergic rhinitis symptoms on patient quality of life (QOL) is currently unknown. OBJECTIVE: In this study, we sought to determine the association between nasal and extranasal symptoms of allergic rhinitis with general health-related QOL. METHODS: A total of 150 participants with ages > 18 years and persistent allergic rhinitis were prospectively recruited for this cross-sectional study. General health-related QOL was measured by using the visual analog scale (VAS) from the five-dimension EuroQol QOL survey (EQ-5D). The severity of nasal and extranasal symptoms was measured by using the 22-item Sino-Nasal Outcome Test (SNOT-22). The severity of nasal obstruction was measured by using the Nasal Obstruction Severity Evaluation (NOSE) scale. Each participant also completed a Rhinitis Control Assessment Test (RCAT). RESULTS: The total SNOT-22 score was significantly correlated with RCAT (r = -0.68 [95% confidence interval {CI}, -0.75 to -0.58]; p < 0.001) and EQ-5D VAS (r = -0.44 [95% CI, -0.56 to -0.30]; p < 0.001). Of the 22 symptoms assessed on the SNOT-22, sleep-related symptoms, and otologic symptoms were associated with the greatest decrease in general health-related QOL. Nasal symptoms were least associated with general health-related QOL. The symptom of nasal obstruction was not at all associated with general health-related QOL (p = 0.267). We confirmed this finding by showing no significant correlation between the NOSE score and EQ-5D VAS (r = -0.05 [95% CI, -0.21 to 0.12]; p = 0.582). CONCLUSION: Sleep and otologic symptoms were associated with the greatest negative impact on QOL in adults with persistent allergic rhinitis and should be routinely assessed in their clinical evaluation.

Facial palsy after dental procedures - Is viral reactivation responsible?

OBJECTIVES: Herpes labialis viral reactivation has been reported following dental procedures, but the incidence, characteristics and outcomes of delayed peripheral facial nerve palsy following dental work is poorly understood. Herein we describe the unique features of delayed facial paresis following dental procedures. MATERIALS AND METHODS: An institutional retrospective review was performed to identify patients diagnosed with delayed facial nerve palsy within 30 days of dental manipulation. Demographics, prodromal signs and symptoms, initial medical treatment and outcomes were assessed. RESULTS: Of 2471 patients with facial palsy, 16 (0.7%) had delayed facial paresis following ipsilateral dental procedures. Average age at presentation was 44 yrs and 56% (9/16) were female. Clinical evaluation was consistent with Bell's palsy in 14 (88%) and Ramsay-Hunt syndrome in 2 patients (12%). Patients developed facial paresis an average of 3.9 days after the dental procedure, with all individuals developing a flaccid paralysis (House Brackmann (HB) grade VI) during the acute stage. 50% of patients developed persistent facial palsy in the form of non-flaccid facial paralysis (HBIII-IV). CONCLUSION: Facial palsy, like herpes labialis, can occur in the days following dental procedures and may also be related to viral reactivation. In this small cohort, long-term facial outcomes appear worse than for spontaneous Bell's palsy.

Steroid use in Lyme disease-associated facial palsy is associated with worse long-term outcomes.

OBJECTIVE: The purpose of this study was to determine whether differences in long-term facial function outcomes following acute Lyme disease-associated facial palsy (LDFP) exist between patients who received antibiotic monotherapy (MT); dual therapy (DT) with antibiotics and corticosteroids; and triple therapy (TT) with antibiotics, corticosteroids, and antivirals. STUDY DESIGN: Retrospective cohort. METHODS: All patients with a prior diagnosis of unilateral LDFP who presented to our center between 2002 and 2015 were retrospectively assessed for inclusion. Two blinded experts graded static, dynamic, and synkinesis parameters of facial functions using standardized video documentation of facial function. RESULTS: Fifty-one patients were included. The mean time of assessment following LDFP onset was 15.1 months (range 0.3-84 months). Significantly worse facial outcomes were seen among those who received DT and TT as compared to those who received MT, most pronounced among those assessed 12 months or later following onset of LDFP (Dynamic-P = 0.031, post hoc MT vs. TT: mean difference [MD], 15.83; 95% confidence interval [CI], 1.54-30.13; P = 0.030. Synkinesis-P = 0.026, post hoc MT vs. DT: MD, 21.50; 95% CI, 0.68-42.32; P = 0.043, post hoc MT vs. TT: MD, 19.22; 95% CI, 2.23-36.22; P = 0.027). CONCLUSION: An association between corticosteroid use in acute LDFP and worse long-term facial function outcomes has been demonstrated. Care should be taken in differentiating viral or idiopathic facial palsy (e.g., Bell palsy) from LDFP. LEVEL OF EVIDENCE: 4. Laryngoscope, 127:1451-1458, 2017.

Bilateral Facial Paralysis: A 13-Year Experience.

BACKGROUND: Bilateral facial palsy is a rare clinical entity caused by myriad disparate conditions requiring different treatment paradigms. Lyme disease, Guillain-Barré syndrome, and leukemia are several examples. In this article, the authors describe the cause, the initial diagnostic approach, and the management of long-term sequelae of bilateral paralysis that has evolved in the authors' center over the past 13 years. METHODS: A chart review was performed to identify all patients diagnosed with bilateral paralysis at the authors' center between January of 2002 and January of 2015. Demographics, signs and symptoms, diagnosis, initial medical treatment, interventions for facial reanimation, and outcomes were reviewed. RESULTS: Of the 2471 patients seen at the authors' center, 68 patients (3 percent) with bilateral facial paralysis were identified. Ten patients (15 percent) presented with bilateral facial paralysis caused by Lyme disease, nine (13 percent) with Möbius syndrome, nine (13 percent) with neurofibromatosis type 2, five (7 percent) with bilateral facial palsy caused by brain tumor, four (6 percent) with Melkersson-Rosenthal syndrome, three (4 percent) with bilateral temporal bone fractures, two (3 percent) with Guillain-Barré syndrome, one (2 percent) with central nervous system lymphoma, one (2 percent) with human immunodeficiency virus infection, and 24 (35 percent) with presumed Bell palsy. Treatment included pharmacologic therapy, physical therapy, chemodenervation, and surgical interventions. CONCLUSIONS: Bilateral facial palsy is a rare medical condition, and treatment often requires a multidisciplinary approach. The authors outline diagnostic and therapeutic algorithms of a tertiary care center to provide clinicians with a systematic approach to managing these complicated patients.

Emerging vs Time-Tested Methods of Facial Grading Among Patients With Facial Paralysis.

IMPORTANCE: Most rehabilitation specialists and many facial reanimation surgeons use the Sunnybrook Facial Grading System (FGS) to measure and detect changes in facial function. The eFACE, an electronic and digitally graded facial measurement scale, was recently created to provide similar information to the Sunnybrook FGS, but with scaling uniformity across all categories of facial function, graphical outputs, and easy-to-use visual analog scales. OBJECTIVES: To establish the correlation between the scores on the eFACE and the Sunnybrook FGS among patients with facial paralysis and to compare the reliability of the 2 scales. DESIGN, SETTING, AND PARTICIPANTS: A retrospective review of medical records identified 109 patients who were evaluated at a facial nerve center by physical therapists using the eFACE and the Sunnybrook FGS on the same day, between November 1, 2014, and May 31, 2015. The level of facial function predicted using the 2 scales was compared to study correlation between the scales. Data analysis was conducted from June 1 to September 1, 2015. MAIN OUTCOMES AND MEASURES: Correlation between the Sunnybrook FGS and the eFACE grading scale. METHODS: Two independent physical therapists evaluated patients using both the eFACE and the Sunnybrook FGS. Scores were compared and the Spearman rank correlation coefficient was calculated between the total scores and each of the 3 subscores, including static, dynamic, and synkinesis scores. The total Sunnybrook FGS synkinesis score (worst score, 15; perfect score, 0) and static score (worst score, 20; perfect score, 0) were normalized to a 100-point scale with the eFACE (perfect score, 100; worst score, 1). RESULTS: eFACE scores ranged from 48 to 100, and Sunnybrook FGS scores ranged from 0 to 100. Among 109 patients, there was a moderately strong correlation between eFACE and Sunnybrook FGS scores in both total and subcategory scores. The Spearman rank correlation coefficient of the total eFACE and Sunnybrook FGS was 0.75 (r < 0.0001). For the static scores, the correlation coefficient was -0.71 (r < 0.0001). The correlation coefficients for the dynamic and synkinesis scores were 0.77 (r < 0.0001) and -0.78 (r < 0.0001), respectively. CONCLUSIONS AND RELEVANCE: There is moderately good agreement between the Sunnybrook FGS and the eFACE. Given the ease of using the eFACE on mobile devices, as well as its additional functionality, it may represent a reasonable facial grading option across disciplines in the future. LEVEL OF EVIDENCE: NA.

[Small red sports on the feet and diarrhoea with 63]. / Eine 63-jährige Patientin mit Petechien an den Füssen und Diarrhö

We report a case of a 63 year old woman with progressive small red spots that can not be diminished with pressure. They appeared on the feet first and spread out to the legs, trunk and arms. This was accompanied with non-bloody diarrhoea and the laboratory diagnostics showed minor impairment of the kidney. The biopsy of the skin showed deposits of immunoglobulin A and C3 complement in the stratum papillare of the small vessels and a necrosis of the wall, which made the diagnosis of the Schönlein-Henoch purpura clear. It was accompanied by a systemic participation of the gastrointestinal tract. The quick reversible renal insufficiency was rather interpreted as a prerenal cause, than as an origin of the Schönlein-Henoch purpura. We further discussed the diagnostic methods, the aetiology and the possible therapy options.