Sclera Image CNN Screening for Early Jaundice Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Patients with jaundice often ignore or misinterpret their symptoms due to a yellowish complexion, leading to delayed diagnosis and organ degeneration, necessitating an effective method to determine jaundice through sclera images.
Innovation Solution
A jaundice analysis system utilizing a database and processing device with a data processing module and deep learning module to train a convolutional neural network using sclera images, employing techniques like cutting, mirroring, and de-reflection processing to generate training data, and using transfer learning to determine bilirubin concentration ranges.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If patients with jaundice symptoms rely on self-observation of their yellowish complexion, then they may mistake their jaundice symptoms for a tanned complexion or ignore their diseases, but this leads to delayed diagnosis and organ degeneration
Solution Approach 1:
The patent introduces an intermediary system consisting of image capture devices and deep learning algorithms that mediate between the patient's physical symptoms and the diagnosis. The system captures images of the sclera and skin, processes them through trained neural networks, and provides objective jaundice detection, eliminating the need for patients to self-diagnose and reducing diagnosis delay
Solution Approach 2:
The patent replaces the mechanical/physical process of human visual inspection and subjective judgment with an automated image processing system. The deep learning model automatically analyzes image features and determines jaundice presence and severity, substituting human perception with algorithmic analysis for more accurate and timely diagnosis
2Measurement precision
If a deep learning model is trained from scratch on jaundice detection data, then the model can be highly specialized for this task, but the training time and computational resources required increase significantly
Solution Approach 1:
The patent applies preliminary action by using pre-trained convolutional neural network models that have already learned general image features from large datasets. This preliminary training is transferred to the jaundice detection task, allowing the model to achieve high detection accuracy without requiring extensive training from scratch, thus reducing training time and computational resources
Solution Approach 2:
The patent changes the parameters of the neural network by fine-tuning the pre-trained model with jaundice-specific data. Instead of training all parameters from scratch, the system adjusts only the necessary parameters for jaundice detection while keeping other parameters fixed or lightly trained, achieving high accuracy with reduced training time
Data Source
AI summary
A jaundice analysis system includes a database and a processing device for accessing the database. The processing device includes: a data processing module for generating a training data according to an image data, correlating the training data with a category data, and storing the training data in the database; and a deep learning module for training a target convolutional neural network module with the training data correlating with the category data to obtain a trained convolutional neural network module. The image data includes a first sclera image. The trained convolutional neural network module of the processing device generates a testing data according to an input image data. The input image data includes a second sclera image of a target subject. The testing data indicates the target subject's bilirubin concentration range.