Writing text with eye gaze only is an appealing hands-free text entry method. However, existing gaze-based text entry methods introduce eye fatigue and are slow in typing speed because they often require users to dwell on letters of a word, or mark the starting and ending positions of a gaze path with extra operations for entering a word. In this paper, we propose GlanceWriter, a text entry method that allows users to enter text by glancing over keys one by one without any need to dwell on any keys or specify the starting and ending positions of a gaze path when typing a word. To achieve so, GlanceWriter probabilistically determines the letters to be typed based on the dynamics of gaze movements and gaze locations. Our user studies demonstrate that GlanceWriter significantly improves the text entry performance over EyeSwipe, a dwell-free input method using “reverse crossing” to identify the starting and ending keys. GlanceWriter also outperforms the dwell-free gaze input method of Tobii’s Communicator 5, a commercial eye gaze-based communication system. Overall, GlanceWriter achieves dwell-free and crossing-free text entry by probabilistically decoding gaze paths, offering a promising gaze-based text entry method.

Text correction on mobile devices usually requires precise and repetitive manual control. In this paper, we present EyeSayCorrect, an eye gaze and voice based hands-free text correction method for mobile devices. To correct text with EyeSayCorrect, the user first utilizes the gaze location on the screen to select a word, then speaks the new phrase. EyeSayCorrect would then infer the user’s correction intention based on the inputs and the text context. We used a Bayesian approach for determining the selected word given an eye-gaze trajectory. Given each sampling point in an eye-gaze trajectory, the posterior probability of selecting a word is calculated and accumulated. The target word would be selected when its accumulated interest is larger than a threshold. The misspelt words have higher priors. Our user studies showed that using priors for misspelt words reduced the task completion time up to 23.79% and the text selection time up to 40.35%, and EyeSayCorrect is a feasible hands-free text correction method on mobile devices.

OBJECTIVE: The increasing translation of artificial intelligence (AI)/machine learning (ML) models into clinical practice brings an increased risk of direct harm from modeling bias; however, bias remains incompletely measured in many medical AI applications. This article aims to provide a framework for objective evaluation of medical AI from multiple aspects, focusing on binary classification models. MATERIALS AND METHODS: Using data from over 56 000 Mass General Brigham (MGB) patients with confirmed severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), we evaluate unrecognized bias in 4 AI models developed during the early months of the pandemic in Boston, Massachusetts that predict risks of hospital admission, ICU admission, mechanical ventilation, and death after a SARS-CoV-2 infection purely based on their pre-infection longitudinal medical records. Models were evaluated both retrospectively and prospectively using model-level metrics of discrimination, accuracy, and reliability, and a novel individual-level metric for error. RESULTS: We found inconsistent instances of model-level bias in the prediction models. From an individual-level aspect, however, we found most all models performing with slightly higher error rates for older patients. DISCUSSION: While a model can be biased against certain protected groups (ie, perform worse) in certain tasks, it can be at the same time biased towards another protected group (ie, perform better). As such, current bias evaluation studies may lack a full depiction of the variable effects of a model on its subpopulations. CONCLUSION: Only a holistic evaluation, a diligent search for unrecognized bias, can provide enough information for an unbiased judgment of AI bias that can invigorate follow-up investigations on identifying the underlying roots of bias and ultimately make a change.

Selecting targets accurately and quickly with eye-gaze input remains an open research question. In this paper, we introduce BayesGaze, a Bayesian approach of determining the selected target given an eye-gaze trajectory. This approach views each sampling point in an eye-gaze trajectory as a signal for selecting a target. It then uses the Bayes' theorem to calculate the posterior probability of selecting a target given a sampling point, and accumulates the posterior probabilities weighted by sampling interval to determine the selected target. The selection results are fed back to update the prior distribution of targets, which is modeled by a categorical distribution. Our investigation shows that BayesGaze improves target selection accuracy and speed over a dwell-based selection method, and the Center of Gravity Mapping (CM) method. Our research shows that both accumulating posterior and incorporating the prior are effective in improving the performance of eye-gaze based target selection.

OBJECTIVE: The United States is experiencing an opioid epidemic. In recent years, there were more than 10 million opioid misusers aged 12 years or older annually. Identifying patients at high risk of opioid use disorder (OUD) can help to make early clinical interventions to reduce the risk of OUD. Our goal is to develop and evaluate models to predict OUD for patients on opioid medications using electronic health records and deep learning methods. The resulting models help us to better understand OUD, providing new insights on the opioid epidemic. Further, these models provide a foundation for clinical tools to predict OUD before it occurs, permitting early interventions. METHODS: Electronic health records of patients who have been prescribed with medications containing active opioid ingredients were extracted from Cerner's Health Facts database for encounters between January 1, 2008, and December 31, 2017. Long short-term memory models were applied to predict OUD risk based on five recent prior encounters before the target encounter and compared with logistic regression, random forest, decision tree, and dense neural network. Prediction performance was assessed using F1 score, precision, recall, and area under the receiver-operating characteristic curve. RESULTS: The long short-term memory (LSTM) model provided promising prediction results which outperformed other methods, with an F1 score of 0.8023 (about 0.016 higher than dense neural network (DNN)) and an area under the receiver-operating characteristic curve (AUROC) of 0.9369 (about 0.145 higher than DNN). CONCLUSIONS: LSTM-based sequential deep learning models can accurately predict OUD using a patient's history of electronic health records, with minimal prior domain knowledge. This tool has the potential to improve clinical decision support for early intervention and prevention to combat the opioid epidemic.

Increased availability of large amounts of address data provides opportunities for data driven studies to improve decision making in business applications and support precision public health with high resolution geolocations. Geocoding large number of addresses is challenging due to high cost and often disclosure of sensitive data to vendors over the Web. Most geocoders take advantage of Web APIs which require sending private addresses over the Internet, which may not be an option for many applications with sensitive data including public health and geo-medicine. Meanwhile, the cost for geocoding massive number of addresses could be high and becomes a major hurdle for many users. To overcome these challenges, we developed an open source on-premise geocoding software EaserGeocoder, which uses a novel integrative geocoding model to achieve high accuracy through integrating multiple open data sources. EaserGeocoder takes advantage of machine learning based approaches to determine best answers from multiple data sources. EaserGeocoder can also be easily parallelized to achieve high scalability through parallelized search and distributed computing. EaserGeocoder is on a par with commercial geocoding systems, outperforms open source systems, and is available for free.

With increased accessibility of large scale open data, public health studies are able to take advantage of integrative spatial big data to increase the spatial resolution to community or neighborhood level. One critical information for such studies is the large number of addresses of patients, which is private and highly sensitive. Geocoding such massive private addresses poses major challenges for public health researchers. Many geocoders provide only Web APIs which require sending private addresses over the Internet, which is not feasible. Commercial geocoders require high licensing fee and often have limitations on daily usage, which becomes a major hurdle for researchers. Scalability is another major challenge for large scale address dataset. In this paper, we present EaserGeocoder, a novel open source geocoder for effectively geocoding massive address datasets. EaserGeocoder takes an integrative approach by using multiple references based on open address data sources contributed by governments or communities. It takes a machine learning approach to automatically find the best answer from candidates produced by multiple references. The system provides high scalability through parallel processing. Our comparative studies demonstrate Easer-Geocoder outperforms open source geocoders and is comparable to commercial ones in terms of both accuracy and error. It provides a cost-effective and feasible solution for large scale public health studies.