Automated Coding Model Refinement via Categorical Crossentropy Loss
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated coding models for medical charts face challenges in accurately predicting secondary diagnoses due to the large number of candidate codes, leading to reduced confidence and adoption, partly because of the complexity in designing appropriate loss functions for deep learning models.
Innovation Solution
The method refines an automated coding model by determining categorical crossentropy loss based on predicted probabilities for candidate codes, allowing for the adjustment of model parameters to improve prediction accuracy, which involves predicting probabilities for each candidate secondary diagnosis, selecting the most probable ones, removing them from the set, and iteratively refining the model to minimize loss.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional gradient descent optimization with backpropagation is used to train deep learning models, then the model can generate predictions, but the accuracy of secondary diagnosis prediction is reduced due to the large number of candidate codes and difficulty in designing appropriate loss functions
Solution Approach 1:
The patent changes the parameter representation from traditional gradient-based optimization to categorical cross-entropy loss calculation. By transforming the loss function to operate on categorical probabilities directly rather than through complex gradient descent mechanisms, the system simplifies the optimization process while improving prediction accuracy for secondary diagnoses among tens of thousands of candidate codes
Solution Approach 2:
The patent extracts the essential predictive signal by focusing on categorical cross-entropy loss calculation from the complex training process. By isolating and emphasizing the categorical probability assessment component, the system removes unnecessary complexity from the loss function design while maintaining effective model training for medical code prediction
2Adaptability or versatility
If the number of candidate secondary diagnoses is increased to tens of thousands of codes, then the comprehensiveness of diagnosis coverage is improved, but the difficulty of extracting particular secondary diagnoses increases, leading to reduced model accuracy
Solution Approach 1:
The patent segments the large set of tens of thousands of candidate secondary diagnoses into manageable categorical groups. By organizing codes into structured categories and applying categorical cross-entropy loss at the category level rather than individually for each code, the system maintains comprehensive diagnosis coverage while improving extraction accuracy through hierarchical organization
Solution Approach 2:
The patent introduces a categorical dimension to organize the large space of secondary diagnoses. By mapping tens of thousands of individual codes into a smaller number of categorical groups and performing prediction at the categorical level, the system reduces the effective search space while maintaining the ability to identify specific diagnoses through category membership
3Ease of manufacture
If traditional loss functions are used for training, then the training process is simpler, but the prediction accuracy for secondary diagnoses is reduced due to inability to handle large candidate code sets effectively
Solution Approach 1:
The patent changes the fundamental parameter of the loss function from traditional gradient-based metrics to categorical cross-entropy. This parameter change maintains relative simplicity in the training process while dramatically improving the model's ability to handle large candidate code sets by directly optimizing categorical probability distributions
Data Source
AI summary
A method, apparatus and computer program product refine an automated coding model, such as for a medical chart. For each respective candidate code from a set of candidate codes, the method predicts a probability of the respective code being contained in a medical chart. The method also selects one of the candidate codes as being contained in the medical chart based upon the probability and removes the selected candidate code from the set of candidate codes. The method then repeatedly predicts the probability of a respective code being contained in the medical chart, selects one of the candidate codes based upon the predicted probability and removes the selected candidate code from the set of candidate codes. The method further determines a categorical crossentropy loss as to permit adjustment of one or more parameters of the automated coding model.