Depression Diagnosis Model Training Data Filtering

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing technologies may misclassify individuals with depressive symptoms who have a Hamilton Depression Scale (HAMD) score of 7 points or less as healthy, due to labeling and machine learning processes.

Innovation Solution

A depressive symptom determination model is trained using feature vectors from conversations of subjects with predetermined depression evaluation scale scores, excluding those with manic-depressive disorders, to accurately identify depressive symptoms regardless of HAMD score.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the estimation model is trained using HAMD-17 scores as labels, then it can distinguish between healthy persons and depressed patients based on depression evaluation, but it may misclassify subjects with depressive symptoms who have low HAMD scores as healthy persons

Engineering Contradiction:

Improvedepression diagnosis accuracyVSAvoiddiagnostic reliability for low-score subjects

Core Design Contradiction:

Measurement precisionVSReliability

Solution Approach 1:

The patent extracts and removes subjects with manic-depressive disorders from the training dataset. By identifying and excluding these subjects based on specific criteria (manic-depressive diagnosis or high manic-depressive evaluation scale scores), the training data becomes more纯净, allowing the model to learn genuine depression patterns without being confounded by bipolar disorder presentations, thereby improving diagnostic reliability for subjects with low HAMD scores

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent performs preliminary filtering and classification of training subjects before model training. By pre-identifying and removing subjects with manic-depressive disorders using extraction conditions and exclusion conditions, the data preparation process ensures that only appropriate subjects contribute to training the depression detection model, preventing future misclassifications

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentEP4570189A1Depression symptom determination device, determination model generation device, and training data generation method

Publication Date: 2025.06.18 FRONTEO INC

Figure 1~2
Figure 3~4
Figure 5~7

AI summary

A depressive symptom determination unit 13 configured to determine a depressive symptom of a subject by inputting a feature vector generated based on a feature quantity of a conversation conducted by a determination target subject to a machine-trained determination model is provided, and determination is performed by a determination model generated by machine learning using, as training data, conversation data of a subject satisfying a predetermined extraction condition and exclusion condition with regard to the depressive symptom. By setting a condition for excluding a subject diagnosed with manic-depressive and a subject whose predetermined manic-depressive evaluation scale score is greater than or equal to a manic-depressive threshold as an exclusion condition, a determination model is machine-trained without being affected by conversation data when manic-depressive disorder is temporarily in a depressive state or manic state, making it possible to determine a depressive symptom of a subject having a non-transient depressive symptom as a characteristic of that person.