Related Experiment Videos
Using an artificial neural network to diagnose hepatic masses
1Department of Health Informatics, University of Tennessee, Memphis 38163.
Journal of Medical Systems
|October 1, 1992
Summary
A neural network was developed to aid in diagnosing hepatic masses using ultrasonography and lab data. This AI achieved 75% accuracy, outperforming radiology residents and offering a valuable tool for liver cancer diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Hepatology
Background:
- Diagnosing hepatic masses is challenging for radiologists.
- Accurate differential diagnosis of liver lesions is crucial for patient outcomes.
- Computerized decision support can assist in complex medical diagnoses.
Purpose of the Study:
- To develop and evaluate a back-propagation neural network for diagnosing five types of hepatic masses.
- To assess the diagnostic accuracy of the neural network compared to human experts.
- To explore the potential of artificial intelligence in improving liver mass analysis.
Main Methods:
- A back-propagation neural network was designed with 35 input features from ultrasonographic and laboratory data.
- The network architecture included an input layer, two hidden layers, and an output layer for five mass classifications.
- The network was trained to a 1% learning tolerance.
Main Results:
- The neural network correctly classified 48 out of 64 hepatic mass cases, achieving 75% accuracy.
- This accuracy surpasses the 50% scored by average radiology residents.
- The AI's performance is below board-certified radiologists (90%) but shows significant potential.
Conclusions:
- A neural network can be a valuable tool for assisting in the differential diagnosis of hepatic masses.
- AI-powered systems show promise in enhancing the analysis of medical imaging for liver conditions.
- Further development of sophisticated neural networks may significantly improve radiographic analysis in clinical practice.