This proof-of-concept study found that smartwatch photoplethysmography coupled with a deep neural network can passively detect AF but with some loss of sensitivity and specificity against a criterion-standard ECG. Further studies will help identify the optimal role for smartwatch-guided rhythm assessment.

Patients exposed to a surgical safety checklist experience better postoperative outcomes, but this could simply reflect wider quality of care in hospitals where checklist use is routine.

We train and validate a semi-supervised, multi-task LSTM on 57,675 person-weeks of data from off-the-shelf wearable heart rate sensors, showing high accuracy at detecting multiple medical conditions, including diabetes (0.8451), high cholesterol (0.7441), high blood pressure (0.8086), and sleep apnea (0.8298). We compare two semi-supervised training methods, semi-supervised sequence learning and heuristic pretraining, and show they outperform hand-engineered biomarkers from the medical literature. We believe our work suggests a new approach to patient risk stratification based on cardiovascular risk scores derived from popular wearables such as Fitbit, Apple Watch, or Android Wear.

No abstract

Background The Clavien–Dindo classification is perhaps the most widely used approach for reporting postoperative complications in clinical trials. This system classifies complication severity by the treatment provided. However, it is unclear whether the Clavien–Dindo system can be used internationally in studies across differing healthcare systems in high‐ (HICs) and low‐ and middle‐income countries (LMICs). Methods This was a secondary analysis of the International Surgical Outcomes Study (ISOS), a prospective observational cohort study of elective surgery in adults. Data collection occurred over a 7‐day period. Severity of complications was graded using Clavien–Dindo and the simpler ISOS grading (mild, moderate or severe, based on guided investigator judgement). Severity grading was compared using the intraclass correlation coefficient (ICC). Data are presented as frequencies and ICC values (with 95 per cent c.i.). The analysis was stratified by income status of the country, comparing HICs with LMICs. Results A total of 44 814 patients were recruited from 474 hospitals in 27 countries (19 HICs and 8 LMICs). Some 7508 patients (16·8 per cent) experienced at least one postoperative complication, equivalent to 11 664 complications in total. Using the ISOS classification, 5504 of 11 664 complications (47·2 per cent) were graded as mild, 4244 (36·4 per cent) as moderate and 1916 (16·4 per cent) as severe. Using Clavien–Dindo, 6781 of 11 664 complications (58·1 per cent) were graded as I or II, 1740 (14·9 per cent) as III, 2408 (20·6 per cent) as IV and 735 (6·3 per cent) as V. Agreement between classification systems was poor overall (ICC 0·41, 95 per cent c.i. 0·20 to 0·55), and in LMICs (ICC 0·23, 0·05 to 0·38) and HICs (ICC 0·46, 0·25 to 0·59). Conclusion Caution is recommended when using a treatment approach to grade complications in global surgery studies, as this may introduce bias unintentionally.

No abstract

Plant disease has long been one of the major threats to world food security due to reduction in the crop yield and quality. Accurate and precise diagnosis of plant diseases has been a significant challenge. Cost-effective automated computational systems for disease diagnosis would facilitate advancements in agriculture. The objective of this paper is to explore computer vision based Artificial Intelligence method for automating the identification of yellow rust disease and improve the accuracy of plant disease identification. The dataset of 2000 images of wheat leaf were collected in the real life experimental conditions of ICAR-Indian Agricultural Research Institute, New Delhi in the crop season during January-April, 2019. Based on our experiment, we propose a deep learning-based approach to detect healthy leaves and yellow rust infected leaves in the wheat crop. The experiments are implemented in python with PyCharm IDE, utilizing the Keras deep learning library backend with TensorFlow. The proposed model achieves 97.3% testing accuracy and 98.42% as the training accuracy. The accuracy of the developed model can be improved further by training it with larger size of the dataset in future. In future, accuracy of computer vision based AI models can be improved by using the larger size training datasets. Also, these models can be used for providing automatic advisory services to the farmers, thereby, adding much needed assistance to the overloaded extension experts.

No abstract