Developing Deep Learning Models for the Classification of Pediatric Elbow Radiographic Abnormalities of Comparable Performance to Physicians: Strategies for Model Optimization With Small Sized Development Sets

BackgroundTo compare the performance of an AI model based on strategies designed to overcome small sized development sets to pediatric ER physicians at a classification triage task of pediatric elbow radiographs. Methods1,314 pediatric elbow lateral radiographs (mean age: 8.2 years) were retrospectively retrieved, binomially classified based on their annotation as normal or abnormal (with pathology), and randomly partitioned into a development set (993 images), tuning set (109 images), second tuning set (100 images) and test set (112 images). The AI model was trained on the development set and utilized the EfficientNet B1 compound scaling network architecture and online augmentations. Its performance on the test set was compared to a group of five physicians (inter-rater agreement: fair). Statistical analysis: AUC of AI model - DeLong method. Performance of AI model and physician groups - McNemar test. ResultsAccuracy of the model on the test set - 0.804 (95% CI, 0.718 - 0.873), AUROC - 0.872 (95% CI, 0.831 - 0.947). AI model performance compared to the physician group on the test set - sensitivity 0.790 (95% CI 0.684 to 0.895) vs 0.649 (95% CI 0.525 to 0.773), p value 0.088; specificity 0.818 (95% CI 0.716 to 0.920) vs 0.873 (95% CI 0.785 to 0.961), p value 0.439.ConclusionsThe AI model for elbow radiograph triage designed with strategies to optimize performance for a small sized development set showed comparable performance to physicians.

An Outperforming Artificial Intelligence Model to Identify Referable Blepharoptosis for General Practitioners (Preprint)

BACKGROUND Accurate identification and prompt referral for blepharoptosis can be challenging for general practitioners. An artificial intelligence-aided diagnostic tool could underpin decision-making. OBJECTIVE To develop an AI model which accurately identifies referable blepharoptosis automatically and to compare the AI model’s performance to a group of non-ophthalmic physicians. METHODS Retrospective 1,000 single-eye images from tertiary oculoplastic clinics were labeled by three oculoplastic surgeons with ptosis, including true and pseudoptosis, versus healthy eyelid. The VGG (Visual Geometry Group)-16 model was trained for binary classification. The same dataset was used in testing three non-ophthalmic physicians. The Gradient-weighted Class Activation Mapping (Grad-CAM) was applied to visualize the AI model RESULTS The VGG16-based AI model achieved a sensitivity of 92% and a specificity of 88%, compared with the non-ophthalmic physician group, who achieved a mean sensitivity of 72% [Range: 68% - 76%] and a mean specificity of 82.67% [Range: 72% - 88%]. The area under the curve (AUC) of the AI model was 0.987. The Grad-CAM results for ptosis predictions highlighted the area between the upper eyelid margin and central corneal light reflex. CONCLUSIONS The AI model shows better performance than the non-ophthalmic physician group in identifying referable blepharoptosis, including true and pseudoptosis, correctly. Therefore, artificial intelligence-aided tools have the potential to assist in the diagnosis and referral of blepharoptosis for general practitioners.

Validation of an endoscopic flavectomy training model

ABSTRACT Objective: to validate a lumbar spine endoscopic flavectomy simulator using the construct method and to assess the acceptability of the simulator in medical education. Methods: thirty medical students and ten video-assisted surgery experienced orthopedists performed an endoscopic flavectomy procedure in the simulator. Time, look-downs, lost instruments, respect for the stipulated edge of the ligamentum flavum, regularity of the incision, GOALS checklist (Global Operative Assessment of Laparoscopic Skills), and responses to the Likert Scale adapted for this study were analyzed. Results: all variables differed between groups. Procedure time was shorter in the physician group (p < 0.001). Look-downs and instrument losses were seven times greater among students than physicians. Half of the students respected the designated incision limits, compared to 80% of the physicians. In the student group, about 30% of the incisions were regular, compared to 100% in the physician group (p < 0.001). The physicians performed better in all GOALS checklist domains. All the physicians and more than 96% of the students considered the activity enjoyable, and approximately 90% believed that the model was realistic and could contribute to medical education. Conclusions: the simulator could differentiate the groups’ experience level, indicating construct validity, and both groups reported high acceptability.

Clinician Roles in Symptoms and Signs Improvement among Advanced Cancer Patients

Background: We, study and research how to improve terminally ill cancer patients’ symptoms and signs to ease their pain. Furthermore we also seek how the dedicated palliative care service effects on the cancer patients’ results.Methods: We analyzed 334 physician-diagnosed terminally ill cancer patients and categorized them into non-dedicated hospice care group of 234 and dedicated hospice care group of 100.Results: Symptoms improvement of dyspnea, fatigue, drowsiness, and dry mouth during the first week of admission were respectively 298 (89.2%), 25 (7.5%), 204 (61.1%), 76 (22.8%). Signs improvement of myoclonus, respiratory secretion, leg edema, and ascites between admission and a week after were 5 (1.5%), 41 (12.3%), 47 (14.1%), 12 (3.6%). Significant differences between dedicated hospice care physician group and non-dedicated hospice care physician group were shown in drowsiness (67.5% vs 46 %, p < 0.001) and respiratory secretion (15% vs 6%, p < 0.028).Conclusions:In conclusion, terminally ill cancer patients who received palliative or hospice service showed significant improvement in symptoms and illness signs. Moreover, family doctors (dedicated hospice physician group) performed better than oncologist (non-dedicated physician group).

Peer review audit of non-specialist occupational physician reports

Background With declining specialist occupational physician (OP) numbers, there is increasing recognition of the importance of non-specialist physicians in occupational health (OH) service delivery, yet to date, this physician group remains understudied and their competency requirements poorly understood. Aims To evaluate the quality of a sample of non-specialist OH reports and compare these with specialist reports. Methods A retrospective peer review audit of a convenience sample of 200 consecutive non-specialist and specialist OH reports from an Irish OH service using an assessment form based on the modified Sheffield Assessment Instrument for Letters SAIL(OH)1. Results Of the 200 peer reviewed OH reports, 159 (80%) were from non-specialists. For all questions, 87% and above of non-specialist reports were ‘satisfactory’ or ‘above expected’. On the overall assessment, out of 10, the mean non-specialist report score was 6.8 (standard deviation (SD) 3–10) and the specialist score was 7.3 (SD 3–10). Comparatively, non-specialist reports highlighted legal/ethical issues marginally more and adhered slightly better to contractual/ethical/legal boundaries, while specialist reports fared better in addressing manager’s questions, in their structure and clarity and in covering all significant aspects of the case, particularly if the case was complex. Conclusions Our findings demonstrate a high standard of OH report quality in this sample of non-specialist OPs that is consistent across all key OH report components. Potential development areas are also identified that can inform education/training tailored to this physician group and assist in competency standard-setting.

Effect of Intensive Care Provided by Nurse Practitioners for Postoperative Patients: A Retrospective Observational Before-and-after Study

Background Since April 2016, nurse practitioners (NPs) have provided inter-professional team collaborative care for postoperative patients in our intensive care unit (ICU). To assess the effectiveness of NP practice, we compared patient outcomes with and without NP care in our ICU. Methods This retrospective observational cohort study was undertaken at Aichi Medical University Hospital in Japan from 01 April 2015 and 31 March 2017. Mechanically-ventilated postoperative patients who stayed in the ICU for >2 days were eligible for inclusion. We compared the patient outcomes between the period without NP care from April 2015 to March 2016 (physician group) and the period with NP care from April 2016 to May 2017 (NP-physician group). The primary outcome was ICU length of stay. Secondary outcomes were mechanical ventilation days and total hospital length of stay. To further determine the factors associated with ICU length of stay, multiple regression analysis was performed for age, Acute Physiology and Chronic Health Evaluation (APACHE) II score, NP-physician group, and physician group. Results A total of 387 patients were included in the study. Among these patients, 213 were assigned to the NP-physician group and 174 were assigned to the physician group. The NP-physician group had shorter ICU length of stay than the physician group (4.8 ± 4.8 days versus 6.7 ± 10.3 days, p < 0.021). There were no significant differences between the two groups in mechanical ventilation days or total hospital length of stay. In the multiple regression analysis for ICU length of stay, APACHE II score and NP-physician group were associated with ICU length of stay. Specifically, NP-physician group reduced ICU length of stay by 2.6 days ( p = 0.003), and APACHE II score one-point decrease reduced ICU length of stay by 0.2 days ( p = 0.012). Conclusions The present study demonstrated shorter ICU length of stay in patients with NP care and a significant association between NP participation in the ICU team and ICU length of stay.