An artificial intelligent diagnostic system on mobile Android terminals for cholelithiasis by lightweight...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Mobile Computing

Background:

• AI for medical image analysis is limited on mobile devices due to resource constraints.
• Existing AI diagnostic tools often require high computational power, hindering mobile application.

Purpose of the Study:

• To develop a novel AI diagnostic system for cholelithiasis (gallstone) recognition on Android mobile devices.
• To create a lightweight AI model suitable for the limited computational power of mobile terminals.

Main Methods:

• Collected and preprocessed a dataset of CT images of cholelithiasis using histogram equalization.
• Developed a lightweight convolutional neural network (MobileNetV2) for feature extraction and gallstone recognition.
• Implemented the deep learning algorithm using Java and C++ for the Android platform, with offline training on PCs.

Main Results:

• The MobileNetV2 system achieved an accuracy rate of approximately 91% for cholelithiasis recognition.
• The number of parameters was significantly reduced to 4.3M for MobileNetV2, compared to other methods (e.g., 36.1M for SSD, 50.7M for YOLOv2).
• Recognition tasks on mobile devices were completed within 4 seconds.

Conclusions:

• A lightweight AI system for gallstone detection is feasible and effective on Android mobile devices.
• The developed MobileNetV2 model offers a balance of high accuracy and reduced computational requirements for mobile medical diagnostics.
• This work paves the way for deploying AI-powered diagnostic tools directly on mobile platforms.