Methods and systems for monitoring and surveillance of autonomous AI in medical imaging

A non-transitory computer-readable medium comprising machine-executable code that, upon execution by one or more computer processors, implements a method for operating a machine-learning medical imaging screening classifier, the method comprising:

• (a) obtaining the machine-learning medical imaging screening classifier, wherein the machine-learning medical imaging screening classifier is trained to process medical image data of a subject to determine a classification output indicative of a presence or an absence of a health state in the subject;
• (b) obtaining medical image data of a plurality of clinical subjects of a medical imaging clinical site, the plurality of clinical subjects comprising one or more positive subjects having the health state and one or more negative subjects not having the health state;
• (c) applying the machine-learning medical imaging screening classifier to the medical image data of the plurality of subjects to determine a plurality of classification outputs indicative of a presence or an absence of a health state in each of the plurality of clinical subjects;
• (d) determining a set of performance metrics of the machine-learning medical imaging screening classifier, based at least in part on (i) the plurality of classification outputs and (ii) an actual presence or actual absence of the health state for each of the plurality of subjects;
• (e) determining that the machine-learning medical imaging screening classifier is validated when the set of performance metrics meets pre-determined criteria, or determining that the machine-learning medical imaging screening classifier is not validated when the set of performance metrics does not meet the pre-determined criteria;
• and
• (f) operating the machine-learning medical imaging screening classifier in an autonomous operating mode when the machine-learning medical imaging screening classifier is validated, or operating the machine-learning medical imaging screening classifier in a semi-autonomous or non-autonomous operating mode when the machine-learning medical imaging screening classifier is not validated.

A computer system for operating a machine-learning radiological screening classifier, comprising:

• one or more computer processors individually or collectively programmed to:
• (a) obtain the machine-learning medical imaging screening classifier, wherein the machine-learning medical imaging screening classifier is trained to process medical image data of a subject to determine a classification output indicative of a presence or an absence of a health state in the subject;
• (b) obtain medical image data of a plurality of clinical subjects of a medical imaging clinical site, the plurality of clinical subjects comprising one or more positive subjects having the health state and one or more negative subjects not having the health state;
• (c) apply the machine-learning medical imaging screening classifier to the medical image data of the plurality of subjects to determine a plurality of classification outputs indicative of a presence or an absence of a health state in each of the plurality of clinical subjects;
• (d) determine a set of performance metrics of the machine-learning medical imaging screening classifier, based at least in part on (i) the plurality of classification outputs and (ii) an actual presence or actual absence of the health state for each of the plurality of clinical subjects;
• (e) determine that the machine-learning medical imaging screening classifier is validated when the set of performance metrics meets pre-determined criteria, or determining that the machine-learning medical imaging screening classifier is not validated when the set of performance metrics does not meet the pre-determined criteria;
• and
• (f) operate the machine-learning medical imaging screening classifier in an autonomous operating mode when the machine-learning medical imaging screening classifier is validated, or operating the machine-learning medical imaging screening classifier in a semi-autonomous or non-autonomous operating mode when the machine-learning medical imaging screening classifier is not validated.