Privacy-aware data transformations
Techniques for data security are provided. Face data for a patient is received, where the face data is linked to an identifier of the patient. A suggested medical device is determined by processing the face data using a recommendation model. Therapy data stored in a first data store and equipment data stored in a second data store are identified based on the identifier, where the equipment data indicates an actual medical device that was provided to the patient. The suggested medical device, the therapy data, and the equipment data are associated.
1 . A method comprising:
receiving face data for a first patient, wherein the face data is identifying of the first patient and is associated with a first identifier of the first patient;
de-identifying the face data;
determining a suggested medical device by processing at least one of the face data or the de-identified face data using a recommendation model;
identifying, based on the first identifier, therapy data stored in a first data store;
identifying, based on the first identifier, equipment data stored in a second data store, wherein the equipment data indicates an actual medical device that was provided to the first patient;
generating linked de-identified data by linking the suggested medical device, the therapy data, the equipment data, and the de-identified face data in a third data store using the first identifier, wherein the linked de-identified data does not include the face data; and
facilitating access to the linked de-identified data without enabling access to the face data.
2 . The method of claim 1 , wherein:
the de-identified face data comprises one or more facial measurements of the first patient, and wherein the de-identified face data does not uniquely identify the first patient, and
the method further comprises storing the face data in a fourth data store discrete from the first, second, and third data stores.
3 . The method of claim 1 , wherein de-identifying the face data comprises:
determining a set of coordinates for a set of facial landmarks based on the face data; and
normalizing the set of coordinates using a normalization value.
4 . The method of claim 1 , wherein:
the face data comprises a three-dimensional (3D) mesh of a face of the first patient, and
the method further comprises de-identifying the face data prior to identifying the therapy data or the equipment data.
5 . The method of claim 4 , wherein:
de-identifying the face data comprises extracting one or more facial measurements based on the 3D mesh,
the one or more facial measurements do not uniquely identify the first patient, and
the method further comprises:
linking the one or more facial measurements with the suggested medical device, therapy data, and equipment data, wherein the face data is not linked to the therapy data or equipment data.
6 . The method of claim 4 , wherein:
de-identifying the face data comprises generating a morphed 3D mesh based on the 3D mesh, wherein the morphed 3D mesh does not uniquely identify the first patient, and
the method further comprises:
linking the morphed 3D mesh with the suggested medical device, therapy data, and equipment data, wherein the face data is not linked to the therapy data or equipment data.
7 . The method of claim 6 , wherein generating the morphed 3D mesh comprises morphing the 3D mesh based on principal component analysis (PCA) and a set of statistical shape models for human faces.
8 . The method of claim 1 , further comprising:
outputting, via a graphical user interface (GUI), the linked suggested medical device, therapy data, and equipment data.
9 . The method of claim 1 , further comprising:
determining, based on the linked suggested medical device and equipment data, that the suggested medical device and actual medical device do not match; and
generating an alert indicating that the suggested medical device and the actual medical device do not match.
10 . The method of claim 9 , further comprising:
transmitting, to a provider that provided the actual medical device, a request for information relating to why the actual medical device was provided.
11 . The method of claim 9 , further comprising transmitting, to the first patient, an indication that the actual medical device does not match the suggested medical device.
12 . The method of claim 1 , further comprising:
determining, based on the linked therapy data and equipment data, that the first patient has one or more issues with the actual medical device; and
refining the recommendation model based on the one or more issues.
13 . The method of claim 1 , further comprising selecting a first measurement technique, of a plurality of measurement techniques, for generating face data based at least in part on the linked suggested medical device, the therapy data, and the equipment data.
14 . The method of claim 1 , wherein the face data was received in response to transmitting, to the first patient, a request to provide the face data.
15 . The method of claim 1 , wherein the suggested medical device is a continuous positive airway pressure (CPAP) mask.
16 . A method, comprising:
accessing, by a user device, a patient survey via a provided link for a first patient;
providing a set of responses for the patient survey, wherein the set of responses facilitate selection of a mask for the first patient;
generating identifying face data by performing a facial scan of the first patient using one or more cameras of the user device;
generating de-identified face data based on de-identifying the identifying face data;
transmitting the de-identified face data, along with the set of responses, to a mask selection component without providing access to the identifying face data, wherein:
the mask selection component uses the de-identified face data and set of responses to select the mask for the first patient;
linked de-identified data is generated by linking the de-identified face data with therapy data for the first patient and equipment data for the first patient;
the identifying face data is not linked with the therapy data or the equipment data; and
access to the linked de-identified data is facilitated without enabling access to the identifying face data.
17 . The method of claim 16 , wherein:
the de-identified face data comprises one or more facial measurements of the first patient that do not uniquely identify the first patient,
the one or more facial measurements are generated by the user device by processing one or more images, captured by the one or more cameras of the user device, using one or more machine learning models, and
the one or more images are not transmitted to the mask selection component.
18 . The method of claim 16 , the method further comprising:
determining a set of coordinates for a set of facial landmarks based on the face data; and
normalizing the set of coordinates using a normalization value.
19 . The method of claim 16 , wherein:
the identifying face data comprises a three-dimensional (3D) mesh of a face of the first patient.
20 . The method of claim 19 , wherein:
de-identifying the identifying face data comprises extracting one or more facial measurements based on the 3D mesh, and
the de-identified face data correspond to the one or more facial measurements.
21 . The method of claim 19 , wherein:
de-identifying the identifying face data comprises generating a morphed 3D mesh based on the 3D mesh, and
the de-identified face data correspond to the morphed 3D mesh.
22 . A system, comprising:
a user device configured to perform an operation comprising:
collecting, from a first patient, survey data; and
determining, using on one or more imaging devices, face data for the first patient;
a first data store configured to store the survey data and the face data for the first patient;
a second data store configured to store therapy data for the first patient, wherein the therapy data indicates patient compliance with one or more prescribed therapies;
a third data store configured to store equipment data for the first patient, wherein the equipment data indicates an actual facial mask provided to the first patient; and
a computing device configured to perform an operation comprising:
selecting a suggested facial mask based on the face data;
de-identifying the face data, such that the face data does not uniquely identify the first patient;
generating linked de-identified data by linking the de-identified face data, the therapy data, the equipment data, and the suggested facial mask in a fourth data store, wherein the linked de-identified data does not include the face data; and
facilitating access to the linked de-identified data without providing access to the face data.