Enhancing paranasal sinus disease detection with AutoML: efficient AI development and evaluation via magnetic resonance imaging.

PURPOSE: Artificial intelligence (AI) in the form of automated machine learning (AutoML) offers a new potential breakthrough to overcome the barrier of entry for non-technically trained physicians. A Clinical Decision Support System (CDSS) for screening purposes using AutoML could be beneficial to ease the clinical burden in the radiological workflow for paranasal sinus diseases. METHODS: The main target of this work was the usage of automated evaluation of model performance and the feasibility of the Vertex AI image classification model on the Google Cloud AutoML platform to be trained to automatically classify the presence or absence of sinonasal disease. The dataset is a consensus labelled Open Access Series of Imaging Studies (OASIS-3) MRI head dataset by three specialised head and neck consultant radiologists. A total of 1313 unique non-TSE T2w MRI head sessions were used from the OASIS-3 repository. RESULTS: The best-performing image classification model achieved a precision of 0.928. Demonstrating the feasibility and high performance of the Vertex AI image classification model to automatically detect the presence or absence of sinonasal disease on MRI. CONCLUSION: AutoML allows for potential deployment to optimise diagnostic radiology workflows and lay the foundation for further AI research in radiology and otolaryngology. The usage of AutoML could serve as a formal requirement for a feasibility study.

Prediction of Waiting Times in A&E.

Predicting waiting times in A&E is a critical tool for controlling the flow of patients in the department. The most used method (rolling average) does not account for the complex context of the A&E. Using retrospective data of patients visiting an A&E service from 2017 to 2019 (pre-pandemic). An AI-enabled method is used to predict waiting times in this study. A random forest and XGBoost regression methods were trained and tested to predict the time to discharge before the patient arrived at the hospital. When applying the final models to the 68,321 observations and using the complete set of features, the random forest algorithm's performance measurements are RMSE=85.31 and MAE=66.71. The XGBoost model obtained a performance of RMSE=82.66 and MAE=64.31. The approach might be a more dynamic method to predict waiting times.

Digital Transformation of Cancer Care in the Era of Big Data, Artificial Intelligence and Data-Driven Interventions: Navigating the Field.

OBJECTIVES: To navigate the field of digital cancer care and define and discuss key aspects and applications of big data analytics, artificial intelligence (AI), and data-driven interventions. DATA SOURCES: Peer-reviewed scientific publications and expert opinion. CONCLUSION: The digital transformation of cancer care, enabled by big data analytics, AI, and data-driven interventions, presents a significant opportunity to revolutionize the field. An increased understanding of the lifecycle and ethics of data-driven interventions will enhance development of innovative and applicable products to advance digital cancer care services. IMPLICATIONS FOR NURSING PRACTICE: As digital technologies become integrated into cancer care, nurse practitioners and scientists will be required to increase their knowledge and skills to effectively use these tools to the patient's benefit. An enhanced understanding of the core concepts of AI and big data, confident use of digital health platforms, and ability to interpret the outputs of data-driven interventions are key competencies. Nurses in oncology will play a crucial role in patient education around big data and AI, with a focus on addressing any arising questions, concerns, or misconceptions to foster trust in these technologies. Successful integration of data-driven innovations into oncology nursing practice will empower practitioners to deliver more personalized, effective, and evidence-based care.

A Machine Learning Solution to Predict Elective Orthopedic Surgery Case Duration.

We used surgery durations, patient demographic and personnel data taken from the East Kent Hospitals University NHS Foundation Trust (EKHUFT) over a period of 10 years (2010-2019) for a total of 25,352 patients that underwent 15 highest volume elective orthopedic surgeries, to predict future surgery durations for the subset of elective surgeries under consideration. As part of this study, we compared two different ensemble machine learning methods random forest regression (RF) and XGBoost (eXtreme Gradient Boosting) regression. The two models were approximately 5% superior to the existing model used by the hospital scheduling system.

Using Machine Learning to Improve Personalised Prediction: A Data-Driven Approach to Segment and Stratify Populations for Healthcare.

Population Health Management typically relies on subjective decisions to segment and stratify populations. This study combines unsupervised clustering for segmentation and supervised classification, personalised to clusters, for stratification. An increase in cluster homogeneity, sensitivity and positive predictive value was observed compared to an unlinked approach. This analysis demonstrates the potential for a cluster-then-predict methodology to improve and personalise decisions in healthcare systems.

Early Prediction of Neoplasms Using Machine Learning: A Study of Electronic Health Records from the Ministry of National Guard Health Affairs in Saudi Arabia.

The early detection and treatment of neoplasms, and in particular the malignant, can save lives. However, identifying those most at risk of developing neoplasms remains challenging. Electronic Health Records (EHR) provide a rich source of "big" data on large numbers of patients. We hypothesised that in the period preceding a definitive diagnosis, there exists a series of ordered healthcare events captured within EHR data that characterise the onset and progression of neoplasms that can be exploited to predict future neoplasms occurrence. Using data from the EHR of the Ministry of National Guard Health Affairs (MNG-HA), a large healthcare provider in Saudi Arabia, we aimed to discover health event patterns present in EHR data that predict the development of neoplasms in the year prior to diagnosis. After data cleaning, pre-processing, and applying the inclusion and exclusion criteria, 5,466 patients were available for model construction: 1,715 cases and 3,751 controls. Two predictive models were developed (using Decision tree (DT), and Random Forests (RF)). Age, gender, ethnicity, and ICD-10-chapter (broad disease classification) codes as predictor variables and the presence or absence of neoplasms as the output variable. The common factors associated with a diagnosis of neoplasms within one or more years after their occurrence across all the models were: (1) age at neoplasms/event diagnosis; (2) gender; and patient medical history of (3) diseases of the blood and blood-forming organs and certain disorders involving immune mechanisms, and (4) diseases of the genitourinary system. Model performance assessment showed that RF has higher Area Under the Curve (AUC)=0.76 whereas the DT was less complex. This study is a demonstration that EHR data can be used to predict future neoplasm occurrence.

Predicting the risk of cancer in adults using supervised machine learning: a scoping review.

OBJECTIVES: The purpose of this scoping review is to: (1) identify existing supervised machine learning (ML) approaches on the prediction of cancer in asymptomatic adults; (2) to compare the performance of ML models with each other and (3) to identify potential gaps in research. DESIGN: Scoping review using the population, concept and context approach. SEARCH STRATEGY: PubMed search engine was used from inception to 10 November 2020 to identify literature meeting following inclusion criteria: (1) a general adult (≥18 years) population, either sex, asymptomatic (population); (2) any study using ML techniques to derive predictive models for future cancer risk using clinical and/or demographic and/or basic laboratory data (concept) and (3) original research articles conducted in all settings in any region of the world (context). RESULTS: The search returned 627 unique articles, of which 580 articles were excluded because they did not meet the inclusion criteria, were duplicates or were related to benign neoplasm. Full-text reviews were conducted for 47 articles and a final set of 10 articles were included in this scoping review. These 10 very heterogeneous studies used ML to predict future cancer risk in asymptomatic individuals. All studies reported area under the receiver operating characteristics curve (AUC) values as metrics of model performance, but no study reported measures of model calibration. CONCLUSIONS: Research gaps that must be addressed in order to deliver validated ML-based models to assist clinical decision-making include: (1) establishing model generalisability through validation in independent cohorts, including those from low-income and middle-income countries; (2) establishing models for all cancer types; (3) thorough comparisons of ML models with best available clinical tools to ensure transparency of their potential clinical utility; (4) reporting of model calibration performance and (5) comparisons of different methods on the same cohort to reveal important information about model generalisability and performance.

Insights into Pharmacotherapy Management for Parkinson's Disease Patients Using Wearables Activity Data.

We investigate what supervised classification models using clinical and wearables data are best suited to address two important questions about the management of Parkinson's Disease (PD) patients: 1) does a PD patient require pharmacotherapy or not, and 2) whether therapies are having an effect. Currently, patient management is suboptimal due to using subjective patient reported episodes to answer these questions. METHODOLOGY: Clinical and real environment sensor data (memory, tapping, walking) was provided by the mPower study (6805 participants). From the data, we derived relevant clinical scenarios: S1) before vs. after initiating pharmacotherapy, and S2) before vs. after taking medication. For each scenario we designed and tested 6 methods of supervised classification. Precision, Accuracy and Area Under the Curve (AUC) were computed using 10-fold cross-validation. RESULTS: The best classification models were: S1) Decision Trees on Tapping activity data (AUC 0.95, 95% CI 0.05); and S2) K-Nearest Neighbours on Gait data (mean AUC 0.70, 95% CI 0.07, 46% participants with AUC > 0.70). CONCLUSIONS: Automatic patient classification based on sensor activity data can objectively inform PD medication management, with significant potential for improving patient care.

Offline eLearning for undergraduates in health professions: A systematic review of the impact on knowledge, skills, attitudes and satisfaction.

BACKGROUND: The world is short of 7.2 million health-care workers and this figure is growing. The shortage of teachers is even greater, which limits traditional education modes. eLearning may help overcome this training need. Offline eLearning is useful in remote and resource-limited settings with poor internet access. To inform investments in offline eLearning, we need to establish its effectiveness in terms of gaining knowledge and skills, students' satisfaction and attitudes towards eLearning. METHODS: We conducted a systematic review of offline eLearning for students enrolled in undergraduate, health-related university degrees. We included randomised controlled trials that compared offline eLearning to traditional learning or an alternative eLearning method. We searched the major bibliographic databases in August 2013 to identify articles that focused primarily on students' knowledge, skills, satisfaction and attitudes toward eLearning, and health economic information and adverse effects as secondary outcomes. We also searched reference lists of relevant studies. Two reviewers independently extracted data from the included studies. We synthesized the findings using a thematic summary approach. FINDINGS: Forty-nine studies, including 4955 students enrolled in undergraduate medical, dentistry, nursing, psychology, or physical therapy studies, met the inclusion criteria. Eleven of the 33 studies testing knowledge gains found significantly higher gains in the eLearning intervention groups compared to traditional learning, whereas 21 did not detect significant differences or found mixed results. One study did not test for differences. Eight studies detected significantly higher skill gains in the eLearning intervention groups, whilst the other 5 testing skill gains did not detect differences between groups. No study found offline eLearning as inferior. Generally no differences in attitudes or preference of eLearning over traditional learning were observed. No clear trends were found in the comparison of different modes of eLearning. Most of the studies were small and subject to several biases. CONCLUSIONS: Our results suggest that offline eLearning is equivalent and possibly superior to traditional learning regarding knowledge, skills, attitudes and satisfaction. Although a robust conclusion cannot be drawn due to variable quality of the evidence, these results justify further investment into offline eLearning to address the global health care workforce shortage.

XML knowledge database of MRI-derived eye models.

In a model-based approach, MR images were used to build a database of individual eye models. In order to store the features of the specific eye morphology in an extensible, structured and Internet-accessible database, an appropriate XML structure was implemented. A document type definition was developed that managed the data of the correlated feature space and defined associations via training data sets. The classification and retrieval system has been implemented in Java and successfully applied to classify data sets. Classified data were then added to the database. The presented approach can be easily transferred to similar classification implementations.

Feature extraction and supervised classification of MR images to support proton radiation therapy of eye tumors.

Proton therapy has the potential for high-precision radiotherapy of retinal tumors. However, the standardized eye models currently used do not fully account for the patient's individual anatomy. To better exploit the data provided by MR images, a model-based approach was used based on a database of eye models. A face recognition algorithm was advanced to define similarity criteria between the reference image and the actual image. After building a high-dimensional feature vector and using a training data set, the reference model was selected by using the minimum Mahalanobis distance between the image to be classified and the reference images.