Multimodal investigation of speech, text, cognitive and facial video features for characterizing depression with and without medication
A system for remotely determining the potential presence of depression in a user. The system includes a virtual agent that administers one or more tasks to the user. The user performs the tasks and the performance is captured by a camera. The captured audiovisual data is sent to a server that derives objective metrics which are then applied to a classifying algorithm. If certain metrics meet certain thresholds, the system determines that the user is exhibiting depression symptoms. In embodiments, the system can further determine whether a user has been taking depression medication.
1 . A method for detecting a possible presence of depression, comprising:
presenting, by a computing device and via a virtual agent, at least one task for a user to perform;
capturing, by the computing device, audiovisual data of a performance of the at least one task by the user;
streaming, by the computing device, the audiovisual data to at least one remote computing device;
segmenting, by the remote computing device, the audiovisual data;
calculating, by the remote computing device, objective metrics for the user;
applying, by the remote computing device, a classifying algorithm to the objective metrics, the classifying algorithm comprising:
(a) determining, by the remote computing device, whether a metric representative of a linguistic feature of noun usage is less than a first threshold value;
(b) determining, by the remote computing device, whether a metric representative of lip and jaw kinematics is less than a second threshold;
(c) determining, by the remote computing device, whether a metric representative of cognitive executive function, attention or memory is greater than a third threshold;
(d) determining, by the remote computing device, whether a metric representative of tongue movement and speech articulation is greater than a fourth threshold;
determining, by the remote computing device, that all of the first threshold, the second threshold, the third threshold and the fourth threshold have been met; and
in response to determining that all of the first threshold, the second threshold, the third threshold and the fourth threshold have been met, determining, by the remote computing device, that the user exhibits depression symptoms; and
determining, by the remote computing device, whether the user may have depression based on an output of the classifying algorithm.
2 . The method of claim 1 , wherein the objective metrics are derived according to one or more of a speech acoustic domain, a facial domain, a linguistic domain, a motor domain, an emotional domain, and a cognitive domain.
3 . The method of claim 1 , wherein the at least one task comprises at least one of a counting task, a consonant-vowel-consonant reading task, an oral diadochokinesis task, a reading task, a picture description task, a reading sentences task, a spontaneous speech task, a forward-and-backward digit span task, a motor task, an emotion elicitation task, a word recall task, a semantic fluency task and a sequential command task.
4 . The method of claim 1 further comprising, after the step of calculating: merging, by the remote computing device, demographics information with the calculated objective metrics.
5 . The method of claim 1 , further comprising refining the objective metrics following the step of calculating, wherein the step of refining the objective metrics comprises:
removing, by the remote computing device, any objective metrics beyond a first predefined number of standard deviations; and
for remaining objective metrics, recalculating, by the remote computing device, a mean and a standard deviation and removing any remaining objective metrics with values beyond a second predefined number of standard deviations.
6 . The method of claim 5 , wherein the first predefined number of standard deviations comprises five standard deviations and the second predefined number of standard deviations comprises three standard deviations.
7 . The method of claim 1 , wherein the metric representative of linguistic feature comprises a noun-to-verb ratio.
8 . The method of claim 1 , wherein the metric representative of lip and jaw kinematics comprises at least one of an average speed of the jaw center and a maximum speed of the lower lip.
9 . The method of claim 1 , wherein the metric representative of cognitive executive function, attention or memory comprises a digit span forward score.
10 . The method of claim 1 , wherein the metric representative of tongue movement and speech articulation comprises an articulation rate in words per minute.
11 . The method of claim 1 , further comprising:
providing, by the remote computing device, the determination that the user exhibits depression symptoms to the computing device; and
presenting, by the computing device, the output to the user.
12 . The method of claim 1 , further comprising providing, by the remote computing device, the determination that the user exhibits depression symptoms to a healthcare provider.
13 . The method of claim 1 , further comprising determining whether the user is likely taking anti-depressant medication, where determining whether the user is likely taking anti-depressant medication further comprises:
(e) determining, by the remote computing device, whether the metric representative of lip and jaw kinematics is greater than a fifth threshold;
(f) determining, by the remote computing device, whether the metric representative of tongue movement and speech articulation is less than a sixth threshold;
(g) determining, by the remote computing device, whether a metric representative of the user's speech acoustics and spectral information is less than a seventh threshold;
(h) determining, by the remote computing device, whether the metric representative of cognitive executive function, attention or memory is greater than an eighth threshold;
determining, by the remote computing device, that all of the fifth threshold, the sixth threshold, the seventh threshold and the eighth threshold have been met; and
determining, by the remote computing device, that the user is likely taking anti-depressant medication in response to determining that all of the fifth threshold, the sixth threshold, the seventh threshold and the eighth threshold have been met.
14 . The method of claim 13 , wherein the metric representative of the user's speech acoustics and spectral information comprises a harmonics-to-noise ratio.