| CPC G16H 50/20 (2018.01) [G06N 20/00 (2019.01); G16H 10/40 (2018.01); G16H 10/60 (2018.01)] | 11 Claims |
|
1. A computerized method comprising:
obtaining a set of eye data of a patient from a medical practitioner in a computer input form;
acquiring a set of dry eye patient data from a set of well-structured dry eye patient data samples, wherein each sample dry eye patient data comprises a plurality of features;
identifying the plurality of data features in the set of well-structured dry eye patient data samples, wherein the step of identifying the plurality of data features in the set of well-structured dry eye patient data samples further comprises:
identifying one or more specified data features by creating a plurality of descriptive plots that provide an initial assessment of the data distribution, data noise and one or more data outliers,
wherein each sample is annotated by at least a domain expert to generate a set of feature annotations for each sample, and
wherein the set of feature annotations for each sample is included in the plurality of data features;
implementing a data cleaning process on the set of well-structured dry eye patient data samples;
implementing a feature selection on the set of well-structured dry eye patient data samples, wherein the feature selection comprises selecting a subset of relevant features for machine-learning model construction;
providing a specified machine-learning (ML) model;
training the ML model with the set of well-structured dry eye patient data samples;
validating the ML model with the set of well-structured dry eye patient data samples;
providing the set of eye data of the patient to the trained and validated ML model;
with the trained and validated ML model, classifying the set of eye data of the patient as a dry eye category and a dry eye type, and wherein a dry eye classification comprises the dry eye category and the dry eye type;
wherein the specified ML model comprises three support vector machine (SVM) Models: a dry eye model of severity and amp type (MST), a dry eye model of severity (MS), and dry eye model of type (MT),
wherein the dry eye model of severity and amp type (MST) classifies the set of eye data as mild-aqueous, mild-mixed, mild-evaporative, moderate-aqueous, moderate-mixed, moderate-evaporative, severe-aqueous, severe-mixed, and severe-evaporative, wherein an MST model confusion matrix is provided with a threshold of 0.2 and an MST Under Curve-Receiver Operator Characteristic (AUC-ROC) is calculated and used as an MST primary classification performance evaluation metric,
wherein the dry eye model of severity (MS) classifies the set of eye data of the patient as mild, moderate, or severe, wherein an MS model confusion matrix is provided with a threshold of 0.2 and an MS AUC-ROC is calculated and used as an MS primary classification performance evaluation metric,
wherein the dry eye model of type (MT) classifies the set of eye data of the patient as aqueous, mixed, or evaporative, wherein an MT model confusion matrix is provided with a threshold of 0.2 and an MT Under Curve AUC-ROC is calculated and used as an MT primary classification performance evaluation metric,
integrating the MST model classification or the MS model classification and the MT model classification into the dry eye classification based on the MST Under Curve AUC-ROC, or the MS Under Curve AUC-ROC and the MT Under curve AUC-ROC;
recommending a therapy based on the dry eye classification;
providing an ML model's reasoning behind the recommended therapy to a medical practitioner, wherein the recommended therapy is provided in a human-readable format, wherein the ML model's reasoning comprises list of variables affecting the recommended therapy, and
wherein there is a selection between either the MST model classification or the MS model classification and the MT model classification and either option is then integrated into the dry eye classification.
|