Neural Network Architecture for Performing Medical Coding

Abstract

Mechanisms are provided to implement a medical coding engine to perform medical coding using a neural network architecture that leverages hierarchical semantics between medical concepts. The medical coding engine configures a medical coding neural network to comprise an first layer of nodes comprising preferred terminology (PT) nodes, a second layer comprising lowest level terminology (LLT) nodes, and a third layer comprising weighted values for each connection between each PT node and each LLT node forming a PT node/LLT node connection. Responsive to receiving an adverse event from a cognitive system, a PT node is identified in the first layer associated with a citation from the adverse event. One or more LLT nodes are identified from the second layer based on the identification PT node and a weight associated with the PT node/LLT node connection. A medical code associated with each the one or more LLT nodes is then output.

1 . A method, in a data processing system comprising at least one processor and at least one memory, the at least one memory comprising instructions that are executed by the at least one processor to cause the at least one processor to be configured to implement a medical coding engine to perform medical coding using a neural network architecture that leverages hierarchical semantics between medical concepts in medical dictionaries, the method comprising:
 configuring, by the medical coding engine, a medical coding neural network to comprise a plurality of layers of nodes configured to map terminology in the citations in adverse event inputs to a corresponding medical code;   receiving, by the medical coding engine, an adverse event input specifying an adverse event and having an adverse event citation;   identifying, by the medical coding engine, a medical code associated with a first terminology in the adverse event citation at least by processing the first terminology with the medical coding neural network to generate at least one lower level terminology corresponding to the first terminology at least based on machine learned weights associated with connections between nodes in the medical coding neural network; and   outputting, by the medical coding engine, a medical code associated with the at least one lower level terminology.   
     
     
         2 . (canceled) 
     
     
         3 . The method of  claim 1 , wherein the medical coding neural network is non-fully associative neural network such that each first node in a first layer of the medical coding neural network is connected to only a selected subset of second nodes in a second layer of the medical coding neural network. 
     
     
         4 . The method of  claim 3 , wherein configuring the medical coding neural network comprises generating a connection between each first node to corresponding subset of second nodes and setting an initial weight value of the connection to an initialized weight value, and wherein the machine learned weights are machine learned by increasing weights of the connections between first nodes and corresponding subsets of second nodes based on an verification of an association between the first node and second nodes in the corresponding subset of second nodes. 
     
     
         5 . The method of  claim 1 , wherein training the medical coding neural network further comprises using back-propagation to modify weights of the connections to minimize a loss function of the medical coding neural network. 
     
     
         6 . The method of  claim 5 , wherein a verified second node output increases a weight associated with the connection associated with the second node output. 
     
     
         7 . The method of  claim 5 , wherein a false second node output decreases a weight associated with the connection associated with the second node output. 
     
     
         8 . The method of  claim 1 , wherein the medical coding neural network processes the adverse event citation at least by:
 generating, by the medical coding neural network, kernel representations of the terms used in the adverse event citation;   extracting, by medical coding neural network, feature maps from the kernel representations;   generating, by medical coding neural network, a pooled output of the feature maps for each of the kernel representations;   concatenating, by medical coding neural network, the pooled outputs of the feature maps for each of the kernel representations to generate a concatenated vector input; and   processing, by medical coding neural network, the concatenated vector input by the trained medical coding neural network to identify a first node.   
     
     
         9 . A computer program product comprising a computer readable storage medium having a computer readable program stored therein, wherein the computer readable program, when executed on a data processing system, causes the data processing system to implement a medical coding engine to perform medical coding using a neural network architecture that leverages hierarchical semantics between medical concepts in medical dictionaries, and further causes the data processing system to:
 configure, by the medical coding engine, a medical coding neural network to comprise a plurality of layers of nodes configured to map terminology in citations in adverse event inputs to a corresponding medical code;   receive, by the medical coding engine, an adverse event input specifying an adverse event and having an adverse event citation;   identify, by the medical coding engine, a medical code associated with a first terminology in the adverse event citation input at least by processing the first terminology with the medical coding neural network to generate at least one lower level terminology corresponding to the first terminology at least based on machine learned weights associated with connections between nodes in the medical coding neural network; and   output, by the medical coding engine, a medical code associated with the at least one lower level terminology.   
     
     
         10 . (canceled) 
     
     
         11 . The computer program product of  claim 9 , wherein the medical coding neural network is non-fully associative neural network such that each first node in a first layer of the medical coding neural network is connected to only a selected subset of second nodes in a second layer of the medical coding neural network. 
     
     
         12 . The computer program product of  claim 11 , wherein configuring the medical coding neural network comprises generating a connection between each first node to corresponding subset of second nodes and setting an initial weight value of the connection to an initialized weight value, and wherein the machine learned weights are machine learned by increasing weights of the connections between first nodes and corresponding subsets of second nodes based on an verification of an association between the first node and second nodes in the corresponding subset of second nodes. 
     
     
         13 . The computer program product of  claim 9 , wherein training the medical coding neural network further comprises using back-propagation to modify weights of the connections to minimize a loss function of the medical coding neural network. 
     
     
         14 . The computer program product of  claim 13 , wherein a verified second node output increases a weight associated with the connection associated with the second node output and wherein a false second node output decreases a weight associated with the connection associated with the second node output. 
     
     
         15 . A data processing system comprising:
 at least one processor; and   at least one memory coupled to the at least one processor, wherein the at least one memory comprises instructions which, when executed by the at least one processor, cause the at least one processor to implement a medical coding engine to perform medical coding using a neural network architecture that leverages hierarchical semantics between medical concepts in medical dictionaries, and further cause the at least one processor to:   configure, by the medical coding engine, a medical coding neural network to comprise a plurality of layers of nodes configured to map terminology in citations in adverse event inputs to a corresponding medical code;   receive, by the medical coding engine, an adverse event input specifying an adverse event and having an adverse event citation;   identify, by the medical coding engine, a medical code associated with a first terminology in the adverse event citation input at least by processing the first terminology with the medical coding neural network to generate at least one lower level terminology corresponding to the first terminology at least based on machine learned weights associated with connections between nodes in the medical coding neural network; and   output, by the medical coding engine, a medical code associated with the at least one lower level terminology.   
     
     
         16 . (canceled) 
     
     
         17 . The data processing system of  claim 15 , wherein the medical coding neural network is non-fully associative neural network such that each first node in a first layer of the medical coding neural network is connected to only a selected subset of second nodes in a second layer of the medical coding neural network. 
     
     
         18 . The data processing system of  claim 17 , wherein configuring the medical coding neural network comprises generating a connection between each first node to corresponding subset of second nodes and setting an initial weight value of the connection to an initialized weight value, and wherein the machine learned weights are machine learned by increasing weights of the connections between first nodes and corresponding subsets of second nodes based on an verification of an association between the first node and second nodes in the corresponding subset of second nodes. 
     
     
         19 . The data processing system of  claim 15 , wherein training the medical coding neural network further comprises using back-propagation to modify weights of the connections to minimize a loss function of the medical coding neural network. 
     
     
         20 . The data processing system of  claim 19 , wherein a verified second node output increases a weight associated with the connection associated with the second node output and wherein a false second node output decreases a weight associated with the connection associated with the second node output. 
     
     
         21 . The method of  claim 1 , wherein the first nodes are preferred terminology nodes and the second nodes are lowest level terminology nodes. 
     
     
         22 . The method of  claim 1 , wherein the plurality of layers of nodes comprise:
 a first layer of nodes comprising preferred terminology (PT) nodes representing preferred terminologies of a medical coding schema;   a second layer of nodes comprising lowest level terminology (LLT) nodes representing a lowest level terminologies of a medical coding schema; and   a third layer of nodes between the first layer of nodes and the second layer of nodes comprising weighted values for connections between PT nodes in the first layer of nodes and LLT nodes in the second layer of nodes, to thereby form PT node/LLT node connections.   
     
     
         23 . The computer program product of  claim 9 , wherein the plurality of layers of nodes comprise:
 a first layer of nodes comprising preferred terminology (PT) nodes representing preferred terminologies of a medical coding schema;   a second layer of nodes comprising lowest level terminology (LLT) nodes representing a lowest level terminologies of a medical coding schema; and   a third layer of nodes between the first layer of nodes and the second layer of nodes comprising weighted values for connections between PT nodes in the first layer of nodes and LLT nodes in the second layer of nodes, to thereby form PT node/LLT node connections.