Search

Four deep learning models identified pneumothorax, fractures, opacity, and nodule or mass on frontal chest radiographs with similar performance to radiologists.

Purpose:The purpose of this study was to compare the clinical acceptability of testosterone levels, time to treatment, and postinstruction questions/problems between in-person and telehealth injection teaching for adolescents and young adults (AYA) initiating gender-affirming testosterone therapy.Methods:A retrospective cohort study was conducted using electronic medical record data from a large pediatric gender health clinic. Patients who received subcutaneous testosterone prescriptions between March 15, 2018, and March 14, 2022, were included. The study compared patients receiving in-person versus telehealth injection teaching. Data were collected on demographics, testosterone levels at 3 and 6 months, time from prescription to treatment, and post-instruction contacts. Statistical analyses included t-tests, chi-square tests, logistic regression, linear regression, and zero-inflated negative binomial.Results:The study included 278 patients, with 136 (48.9%) receiving in-person teaching and 142 (51.1%) receiving telehealth teaching. There were no significant differences in baseline characteristics between groups. Clinical effectiveness, indicated by testosterone levels at 3 and 6 months, revealed no significant difference between instruction methods (p= 0.768 and p= 0.350). Time to treatment initiation was comparable (in-person: 15.34 days; telehealth: 18.02 days), with no significant difference in adjusted analysis (p= 0.204). Post-instruction contacts were slightly higher in the telehealth group (2.87 vs. 2.42, p= 0.040), but injection-related questions were rare and similar between groups (p= 0.650).Conclusion:Telehealth instruction for testosterone injection is as effective as in-person teaching methods for AYA initiating gender-affirming care. The findings support the continued use of telehealth to enhance access to gender-affirming care, particularly in light of its clinical efficacy and patient acceptability.

Gastrointestinal diseases, such as gastric cancer, are a major cause of deaths and economic losses. One common diagnostic procedure for detecting these diseases is esophagogastroduodenoscopy (EGD), in which a flexible tube with a camera is inserted into the upper gastrointestinal system to examine it. However, the movement of the gastrointestinal tract and the skill of the physician can sometimes result in blind spots, where potential abnormalities may be missed. To address this issue, we develop a deep learning-based system that tracks the camera’s location during EGD to help doctors avoid potential blind spots. Our system can classify twelve different parts of the upper gastrointestinal system, including the background, with an accuracy of 87.22%. Our analysis showed that most misclassifications occurred between adjacent areas, and we explore two methods, attention block and MoCo pre-training, that can improve the classification accuracy by about 2%. When evaluated on annotated EGD surgery videos, our model achieves a precision of 89.6% and a recall of 88.2%. This system has the potential to improve the accuracy of EGD diagnoses and ultimately reduce the impact of gastrointestinal diseases.

To link to full-text access for this article, visit this link: http://dx.doi.org/10.1016/j.jacr.2009.07.003 Byline: Alexander Ding (a), Jason W. Sharp (b) Author Affiliation: (a) Department of Radiology, Massachusetts General Hospital/Harvard Medical School, Boston, Massachusetts (b) Department of Radiology, University of Texas Southwestern Medical Center at Dallas, Dallas, Texas