The Resilience Doc is vision A.I. for mental health diagnostics.

 

Textpert developed technology to observe a video chat via phone or computer and assess your resilience to anxiety and depression.

We wrote a paper* about our 750-person validation study and it was accepted into an APA peer reviewed journal (Psychological Assessment).

The Science

A.I. to detect resilience using vision, audio & NLP models

1_pixelated_neuralNetworkStructure_RD.pn

The Resilience Doc (RD) combines predictions using an ensemble neural network — similar to the neural network structure in the human brain.

Inputs from body language, tonality, & speech are analyzed through hidden layers & nodes. The RD identifies patterns then generates Resilience Scores on a 0-100 scale.

The RD detected resilience to depression and anxiety with 90% sensitivity, 80% accuracy & 90% specificity on blind validated models

1 terabyte

in-house data

750 Participants evaluated

The RD can process 20 seconds of extemporaneous conversation and assess resilience to depression and anxiety. Video chats via computer or smartphone can now create useful mental health data

Additional Findings

Figure 1: Bing sentiment analysis

AiME corpus Bing sentiment analysis

Figure 1 Each word uttered during the study was  assigned as positive or negative. People with higher depression risk respond with more negative words 

Figure 2: Word variance by PHQ-9

AiME corpus PHQ-9 TFIDF analyis

Figure 2 The plot displays the most significant words per PHQ-9 score. Depressed individuals utilize more ordinary language and overuse the word "really"

Figure 3: Audio encoding

audio features.png

Figure 3 The graph represents various inflection frequencies from just one sentence, digitized and encoded for audio processing. The RD observes inflection variances and analyzes patterns 

Figure 4: Word analysis

Figure 4 The RD analyzes overall sentiment and each word is scored across 250 dimensions. The RD observes patterns and identifies the meaning of each statement

sentiment_features.png