A machine learning approach to identifying suicide risk among text-based crisis counseling encounters.

Introduction: With the increasing utilization of text-based suicide crisis counseling, new means of identifying at risk clients must be explored. Natural language processing (NLP) holds promise for evaluating the content of crisis counseling; here we use a data-driven approach to evaluate NLP methods in identifying client suicide risk. Methods: De-identified crisis counseling data from a regional text-based crisis encounter and mobile tipline application were used to evaluate two modeling approaches in classifying client suicide risk levels. A manual evaluation of model errors and system behavior was conducted. Results: The neural model outperformed a term frequency-inverse document frequency (tf-idf) model in the false-negative rate. While 75% of the neural model's false negative encounters had some discussion of suicidality, 62.5% saw a resolution of the client's initial concerns. Similarly, the neural model detected signals of suicidality in 60.6% of false-positive encounters. Discussion: The neural model demonstrated greater sensitivity in the detection of client suicide risk. A manual assessment of errors and model performance reflected these same findings, detecting higher levels of risk in many of the false-positive encounters and lower levels of risk in many of the false negatives. NLP-based models can detect the suicide risk of text-based crisis encounters from the encounter's content.

REFLACX, a dataset of reports and eye-tracking data for localization of abnormalities in chest x-rays.

Deep learning has shown recent success in classifying anomalies in chest x-rays, but datasets are still small compared to natural image datasets. Supervision of abnormality localization has been shown to improve trained models, partially compensating for dataset sizes. However, explicitly labeling these anomalies requires an expert and is very time-consuming. We propose a potentially scalable method for collecting implicit localization data using an eye tracker to capture gaze locations and a microphone to capture a dictation of a report, imitating the setup of a reading room. The resulting REFLACX (Reports and Eye-Tracking Data for Localization of Abnormalities in Chest X-rays) dataset was labeled across five radiologists and contains 3,032 synchronized sets of eye-tracking data and timestamped report transcriptions for 2,616 chest x-rays from the MIMIC-CXR dataset. We also provide auxiliary annotations, including bounding boxes around lungs and heart and validation labels consisting of ellipses localizing abnormalities and image-level labels. Furthermore, a small subset of the data contains readings from all radiologists, allowing for the calculation of inter-rater scores.

SweetPea: A standard language for factorial experimental design.

Experimental design is a key ingredient of reproducible empirical research. Yet, given the increasing complexity of experimental designs, researchers often struggle to implement ones that allow them to measure their variables of interest without confounds. SweetPea ( https://sweetpea-org.github.io/ ) is an open-source declarative language in Python, in which researchers can describe their desired experiment as a set of factors and constraints. The language leverages advances in areas of computer science to sample experiment sequences in an unbiased way. In this article, we provide an overview of SweetPea's capabilities, and demonstrate its application to the design of psychological experiments. Finally, we discuss current limitations of SweetPea, as well as potential applications to other domains of empirical research, such as neuroscience and machine learning.

How do you feel? Using natural language processing to automatically rate emotion in psychotherapy.

Emotional distress is a common reason for seeking psychotherapy, and sharing emotional material is central to the process of psychotherapy. However, systematic research examining patterns of emotional exchange that occur during psychotherapy sessions is often limited in scale. Traditional methods for identifying emotion in psychotherapy rely on labor-intensive observer ratings, client or therapist ratings obtained before or after sessions, or involve manually extracting ratings of emotion from session transcripts using dictionaries of positive and negative words that do not take the context of a sentence into account. However, recent advances in technology in the area of machine learning algorithms, in particular natural language processing, have made it possible for mental health researchers to identify sentiment, or emotion, in therapist-client interactions on a large scale that would be unattainable with more traditional methods. As an attempt to extend prior findings from Tanana et al. (2016), we compared their previous sentiment model with a common dictionary-based psychotherapy model, LIWC, and a new NLP model, BERT. We used the human ratings from a database of 97,497 utterances from psychotherapy to train the BERT model. Our findings revealed that the unigram sentiment model (kappa = 0.31) outperformed LIWC (kappa = 0.25), and ultimately BERT outperformed both models (kappa = 0.48).

Prediction of Obstructive Lung Disease from Chest Radiographs via Deep Learning Trained on Pulmonary Function Data.

Background: Chronic obstructive pulmonary disease (COPD), the third leading cause of death worldwide, is often underdiagnosed. Purpose: To develop machine learning methods to predict COPD using chest radiographs and a convolutional neural network (CNN) trained with near-concurrent pulmonary function test (PFT) data. Comparison is made to natural language processing (NLP) of the associated radiologist text reports. Materials and Methods: This IRB-approved single-institution retrospective study uses 6749 two-view chest radiograph exams (2012-2017, 4436 unique subjects, 54% female, 46% male), same-day associated radiologist text reports, and PFT exams acquired within 180 days. The Image Model (Resnet18 pre-trained with ImageNet CNN) is trained using frontal and lateral radiographs and PFTs with 10% of the subjects for validation and 19% for testing. The NLP Model is trained using radiologist text reports and PFTs. The primary metric of model comparison is the area under the receiver operating characteristic curve (AUC). Results: The Image Model achieves an AUC of 0.814 for prediction of obstructive lung disease (FEV1/FVC <0.7) from chest radiographs and performs better than the NLP Model (AUC 0.704, p<0.001) from radiologist text reports where FEV1 = forced expiratory volume in 1 second and FVC = forced vital capacity. The Image Model performs better for prediction of severe or very severe COPD (FEV1 <0.5) with an AUC of 0.837 versus the NLP model AUC of 0.770 (p<0.001). Conclusion: A CNN Image Model trained on physiologic lung function data (PFTs) can be applied to chest radiographs for quantitative prediction of obstructive lung disease with good accuracy.

Development and Evaluation of ClientBot: Patient-Like Conversational Agent to Train Basic Counseling Skills.

BACKGROUND: Training therapists is both expensive and time-consuming. Degree-based training can require tens of thousands of dollars and hundreds of hours of expert instruction. Counseling skills practice often involves role-plays, standardized patients, or practice with real clients. Performance-based feedback is critical for skill development and expertise, but trainee therapists often receive minimal and subjective feedback, which is distal to their skill practice. OBJECTIVE: In this study, we developed and evaluated a patient-like neural conversational agent, which provides real-time feedback to trainees via chat-based interaction. METHODS: The text-based conversational agent was trained on an archive of 2354 psychotherapy transcripts and provided specific feedback on the use of basic interviewing and counseling skills (ie, open questions and reflections-summary statements of what a client has said). A total of 151 nontherapists were randomized to either (1) immediate feedback on their use of open questions and reflections during practice session with ClientBot or (2) initial education and encouragement on the skills. RESULTS: Participants in the ClientBot condition used 91% (21.4/11.2) more reflections during practice with feedback (P<.001) and 76% (14.1/8) more reflections after feedback was removed (P<.001) relative to the control group. The treatment group used more open questions during training but not after feedback was removed, suggesting that certain skills may not improve with performance-based feedback. Finally, after feedback was removed, the ClientBot group used 31% (32.5/24.7) more listening skills overall (P<.001). CONCLUSIONS: This proof-of-concept study demonstrates that practice and feedback can improve trainee use of basic counseling skills.

NLIZE: A Perturbation-Driven Visual Interrogation Tool for Analyzing and Interpreting Natural Language Inference Models.

With the recent advances in deep learning, neural network models have obtained state-of-the-art performances for many linguistic tasks in natural language processing. However, this rapid progress also brings enormous challenges. The opaque nature of a neural network model leads to hard-to-debug-systems and difficult-to-interpret mechanisms. Here, we introduce a visualization system that, through a tight yet flexible integration between visualization elements and the underlying model, allows a user to interrogate the model by perturbing the input, internal state, and prediction while observing changes in other parts of the pipeline. We use the natural language inference problem as an example to illustrate how a perturbation-driven paradigm can help domain experts assess the potential limitation of a model, probe its inner states, and interpret and form hypotheses about fundamental model mechanisms such as attention.

Visual Exploration of Semantic Relationships in Neural Word Embeddings.

Constructing distributed representations for words through neural language models and using the resulting vector spaces for analysis has become a crucial component of natural language processing (NLP). However, despite their widespread application, little is known about the structure and properties of these spaces. To gain insights into the relationship between words, the NLP community has begun to adapt high-dimensional visualization techniques. In particular, researchers commonly use t-distributed stochastic neighbor embeddings (t-SNE) and principal component analysis (PCA) to create two-dimensional embeddings for assessing the overall structure and exploring linear relationships (e.g., word analogies), respectively. Unfortunately, these techniques often produce mediocre or even misleading results and cannot address domain-specific visualization challenges that are crucial for understanding semantic relationships in word embeddings. Here, we introduce new embedding techniques for visualizing semantic and syntactic analogies, and the corresponding tests to determine whether the resulting views capture salient structures. Additionally, we introduce two novel views for a comprehensive study of analogy relationships. Finally, we augment t-SNE embeddings to convey uncertainty information in order to allow a reliable interpretation. Combined, the different views address a number of domain-specific tasks difficult to solve with existing tools.

A Comparison of Natural Language Processing Methods for Automated Coding of Motivational Interviewing.

Motivational interviewing (MI) is an efficacious treatment for substance use disorders and other problem behaviors. Studies on MI fidelity and mechanisms of change typically use human raters to code therapy sessions, which requires considerable time, training, and financial costs. Natural language processing techniques have recently been utilized for coding MI sessions using machine learning techniques, rather than human coders, and preliminary results have suggested these methods hold promise. The current study extends this previous work by introducing two natural language processing models for automatically coding MI sessions via computer. The two models differ in the way they semantically represent session content, utilizing either 1) simple discrete sentence features (DSF model) and 2) more complex recursive neural networks (RNN model). Utterance- and session-level predictions from these models were compared to ratings provided by human coders using a large sample of MI sessions (N=341 sessions; 78,977 clinician and client talk turns) from 6 MI studies. Results show that the DSF model generally had slightly better performance compared to the RNN model. The DSF model had "good" or higher utterance-level agreement with human coders (Cohen's kappa>0.60) for open and closed questions, affirm, giving information, and follow/neutral (all therapist codes); considerably higher agreement was obtained for session-level indices, and many estimates were competitive with human-to-human agreement. However, there was poor agreement for client change talk, client sustain talk, and therapist MI-inconsistent behaviors. Natural language processing methods provide accurate representations of human derived behavioral codes and could offer substantial improvements to the efficiency and scale in which MI mechanisms of change research and fidelity monitoring are conducted.