System for providing validation of deep learning based prescription efficacy

1 . A system for computing a probable treatment efficacy, for use with a user device, comprising:

a processor;

an accelerator comprising a tensor streaming processor coupled to the processor; and

memory accessible by the processor, wherein computer program instructions are stored in the memory and executable by the processor to perform:

creating a first plurality of embeddings from a first plurality of patient healthcare records using a pretrained deep learning model, having had personally identifiable information removed and having added a first site locator, associated with a first site, wherein the first site locator is encrypted, to each embedding of the first plurality of embeddings;

creating a second plurality of embeddings from a second plurality of patient healthcare records using the pretrained deep learning model, having had personally identifiable information removed and having added a second site locator, associated with a second site, wherein the second site locator is encrypted, to each embedding of the second plurality of embeddings;

combining the first plurality of embeddings and the second plurality of embeddings into a merged database of embeddings;

receiving a new patient healthcare record for a new patient;

creating a new embedding from the new patient healthcare record using the pretrained deep learning model;

calculating a probable efficacy of a selected treatment of a plurality of treatments based on the merged database of embeddings, wherein the calculating is performed using the tensor streaming processor;

determining a recommended treatment, of the plurality of treatments, to be prescribed to the new patient based on the new embedding and on the calculated probable efficacy of the plurality of treatments, wherein the determining is performed using the tensor streaming processor; and

reporting the recommended treatment to the user device.

2 . The system of claim 1 , wherein the first plurality of healthcare records is a copy of an original first plurality of healthcare records at the first site and wherein the second plurality of healthcare records is a copy of an original second plurality of healthcare records at the second site.

3 . The system of claim 2 , wherein the first plurality of healthcare records is deleted after the creation of the first plurality of embeddings and the second plurality of healthcare records is deleted after the creation of the second plurality of embeddings.

4 . The system of claim 1 , wherein calculating the probable efficacy of the selected treatment of the plurality of treatments based on the merged database of embeddings comprises:

capturing a plurality of features from each embedding of the merged database of embeddings; and

training a classification model using a subset of the captured plurality of features of each embedding of the merged database of embeddings to predict the probable efficacy of the selected treatment, wherein the training is performed using the tensor streaming processor.

5 . The system of claim 4 , wherein the subset of the captured plurality of features is based on demographic information.

6 . The system of claim 5 , wherein the demographic information comprises at least one of geographical data, racial data, ethnic data, cultural data, gender-related data, or age-related data.

7 . The system of claim 4 , further comprising an accelerator coupled to the processor to assist in capturing the plurality of features and training the classification model to predict the probable efficacy of the selected treatment wherein the tensor streaming processor comprises one or more additional processors configured to function as a single core to assist the processor.

8 . The system of claim 1 , wherein the processor further performs the steps of:

receiving a site locator of an embedding of the merged database of embeddings most closely associated with the recommended treatment;

requesting additional information about the recommended treatment from a selected site identified by the selected site locator;

receiving the requested additional information from the selected site; and

reporting to the user device the received additional information along with the recommended treatment.

9 . The system of claim 1 , wherein the processor further performs the step of receiving from the user device an acknowledgement when the recommended treatment was administered to the new patient.

10 . The system of claim 1 , wherein the processor further performs the step of receiving from the user device an indication of the efficacy of the recommended treatment.

11 . A compute platform for calculating a probable treatment efficacy by creating a semantic map from a merged database of embeddings, wherein each embedding of the merged database of embeddings is created using a pretrained deep learning model and is associated with a site, an encrypted site locator, and a treatment of a plurality of treatments, the compute platform comprising:

a processor;

an accelerator comprising a tensor streaming processor coupled to the processor; and

memory accessible by the processor, wherein computer program instructions are stored in the memory and executable by the processor to perform:

receiving a new embedding of a new patient's healthcare records from a new site, wherein the new embedding is created using the pretrained deep learning model;

determining a recommended treatment from the plurality of treatments to be recommended to the new patient by identifying a previous embedding of the merged database of embeddings similar to the new embedding, wherein the determining is performed using the tensor streaming processor;

reporting to the new site the recommended treatment of the plurality of treatments and the site locator associated with the previous embedding;

linking the new site to a site associated with the previous embedding based on the site locator associated with the previous embedding; and

requesting that the site associated with the previous embedding provide additional information to the new site to substantiate that the recommended treatment is applicable to the new patient.

12 . The compute platform of claim 11 , wherein identifying the previous embedding of the merged database of embeddings comprises calculating a similarity of a plurality of similarities between the new embedding and each embedding of the merged database of embeddings and identifying the previous embedding as that embedding of the merged database of embeddings which has the greatest similarity of the plurality of similarities, and wherein if no similarity of the plurality of similarities exceeds a pre-defined threshold, then not identifying a previous embedding and reporting the lack of a previous embedding to the new site.

13 . The compute platform of claim 11 , wherein identifying the previous embedding of the new embedding comprises:

capturing a plurality of features from each embedding of the merged database of embeddings;

training a classification model using a subset of the captured plurality of features of each embedding of the merged database of embeddings, wherein the training is performed using the tensor streaming processor;

applying the trained classification model on the new embedding to determine a probability that each treatment of the plurality of treatments will be effective for the new patient; and

identifying as the previous embedding of the merged database of embeddings that embedding of the merged database of embeddings which is associated with the highest probability that the associated treatment will be effective for the new patient.

14 . The compute platform of claim 13 , further comprising an accelerator coupled to the processor to assist in capturing the plurality of features and in training the classification model, wherein the tensor streaming processor comprises one or more additional processors configured to function as a single core to assist the processor.

15 . A system, for use with a user device, for computing a probable treatment efficacy on a merged database of embeddings, wherein each embedding of the merged database of embeddings is created using a pretrained deep learning model and is associated with a site, an encrypted site locator, and a treatment of a plurality of treatments, comprising:

a processor;

memory accessible by the processor, wherein computer program instructions are stored in the memory and executable by the processor to perform:

capturing a plurality of features from each embedding of the merged database of embeddings;

training a classification model using a subset of the captured plurality of features of each embedding of the merged database of embeddings to determine a probable efficacy of a treatment of the plurality of treatments, wherein the training is performed using the tensor streaming processor, and

sending the trained classification model to the user device;

receiving, at the user device, a new embedding associated with a new patient's healthcare records from a new site, wherein the new embedding is created using the pretrained deep learning model;

determining, at the user device, a recommended treatment of the plurality of treatments, to be prescribed to the new patient based on the trained classification model and the new embedding;

determining, at the user device, a recommended site associated with the recommended treatment of the plurality of treatments; and

establishing, at the user device, a connection to the recommended site to retrieve validation information to support the recommended treatment.

16 . The system of claim 15 , wherein receiving the new embedding comprises the steps of:

receiving, at the user device, a copy of the new patient's healthcare records from the new site;

creating, at the user device, the new embedding based on the copy of the new patient's healthcare records, wherein the creating uses the pretrained deep learning model; and

deleting, at the user device, the received copy of the new patient's healthcare records.

17 . A system for creating a merged database of embeddings of patient healthcare records for use with a user device comprising:

a processor;

memory accessible by the processor, wherein computer program instructions are stored in the memory and executable by the processor to perform:

removing personally identifiable information from each patient healthcare record of a first plurality of patient healthcare records from a first site;

removing personally identifiable information from each patient healthcare record of a second plurality of patient healthcare records from a second site;

creating a first plurality of embeddings from the first plurality of patient healthcare records using a pretrained deep learning model and adding a first site locator, associated with the first site, wherein the first site locator is encrypted, to each embedding of the first plurality of embeddings;

creating a second plurality of embeddings from the second plurality of patient healthcare records using the pretrained deep learning model and adding a second site locator, associated with the second site, wherein the second site locator is encrypted, to each embedding of the second plurality of embeddings;

combining the first plurality of embeddings and the second plurality of embeddings into a merged database of embeddings;

capturing a plurality of features from each embedding of the merged database of embeddings;

training a classification model using a subset of the captured plurality of features of each embedding of the merged database of embeddings to determine the probable efficacy of each treatment of the plurality of treatments, wherein the training is performed using the tensor streaming processor; and

transmitting the trained classification model to the user device.

18 . The system of claim 17 , wherein the first plurality of healthcare records is a copy of an original first plurality of healthcare records from the first site and wherein the second plurality of healthcare records is a copy of an original second plurality of healthcare records from the second site.

19 . The system of claim 17 , further comprising, at the user device:

receiving the trained classification model;

receiving a new patient healthcare record;

converting the new patient healthcare record into a new embedding, wherein the converting uses the pretrained deep learning model;

determining a recommended treatment of the plurality of treatments to be prescribed to the new patient based on the trained classification model and the new embedding;

determining a recommended site associated with the recommended treatment of the plurality of treatments;

establishing a connection to the recommended site to retrieve validation information in support of the recommended treatment; and

displaying the recommended treatment and the validation information in support of the recommended treatment.

20 . The system of claim 19 , further comprising transmitting, from the user device to the processor, an acknowledgement when the recommended treatment was administered to the new patient.

21 . The system of claim 19 , further comprising transmitting, from the user device to the processor, an indication of the efficacy of the recommended treatment.