Systems and methods for software design control and quality assurance

Inventors

Gottlieb, Carl • Trees, Jason

Assignees

Akili Interactive Labs Inc

Abstract

A design control architecture and software quality assurance system for analyzing stimulus-response patterns in user activity data. A design control architecture and software quality assurance system may process user activity data according to a classifier model to assess an impact of an incremental design change to one or more quality measures of a software product. In accordance with certain embodiments the one or more quality measures may include one or more efficacy, safety, and/or performance metrics associated with one or more features of the software product. In certain embodiments, the software product may comprise a digital health intervention or software as a medical device product.

Core Innovation

A software quality assurance system presents, with a user computing device via a user interface, an instance of a software build to a user. The software build comprises at least one feature comprising one or more computerized stimuli or interactions configured to elicit an expected stimulus-input pattern from the user in response to the computerized stimuli or interactions, and the expected stimulus-input pattern comprises a clinically validated stimulus-response pattern for treating or targeting one or more neurological, psychological and/or somatic condition in the user.

The system receives, with at least one sensor communicably engaged with the user computing device, a plurality of user inputs in response to the presenting of the one or more computerized stimuli or interactions, wherein the plurality of user inputs comprises user activity data for a session of the software build. The user activity data is communicated from a first server comprising a software development subsystem server to a second server comprising a design control subsystem server via an application programming interface, and is received at a classification module executing on the second server and communicably engaged with a design control subsystem database via a simple notification service or a simple queue service.

The classification module processes the user activity data to determine one or more actual stimulus-input patterns for each user input, where the classification module comprises a computer-implemented machine learning framework comprising an ensemble learning model or a supervised learning model configured to classify one or more variables between the user activity data and the clinically validated stimulus-response pattern. The classification module calculates a degree of conformity between the one or more actual stimulus-input patterns and the clinically validated stimulus-response pattern, calculates at least one output value including a qualitative or quantitative degree of conformity, and calculates a measure of net therapeutic activity within the session.

Based on the at least one output value and the measure of net therapeutic activity within the session, the second server determines a pass/fail status for the software build according to a minimum performance threshold. The minimum performance threshold comprises a minimum degree of conformity between the one or more actual stimulus-input patterns and the clinically validated stimulus-response pattern and a minimum measure of therapeutic activity delivered to the user within the session of the software build.

Claims Coverage

The independent claims cover a method, a system, and a non-transitory computer-readable medium for software quality assurance. Across the independent claims, five inventive features are consistently present: presenting software build stimuli linked to clinically validated stimulus-response patterns, collecting user activity data, communicating the data to a design control subsystem, classifying actual stimulus-input patterns with machine learning, and determining pass/fail status using conformity and therapeutic-activity thresholds.

The independent claims jointly cover generating QA outcomes by presenting software build stimuli linked to clinically validated stimulus-response patterns, collecting user activity data during a session, using a machine learning classification module to determine actual stimulus-input patterns and compute degrees of conformity, and then calculating net therapeutic activity to determine pass/fail status using minimum performance thresholds tied to conformity and therapeutic activity.

Stated Advantages

Documented Applications

No documented applications found