Using neural networks to diagnose cancer

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Medical Imaging Analysis

Background:

• Artificial neural networks (ANNs) are computational models inspired by the brain's structure.
• ANNs have shown increasing success in clinical diagnosis over the past five years.
• These networks learn by processing examples to achieve acceptable results.

Purpose of the Study:

• To develop and evaluate a prototype artificial neural network for diagnosing renal conditions using ultrasound data.
• To assess the accuracy of the ANN in differentiating renal cell carcinoma from renal cysts and other conditions.

Main Methods:

• A nonlinear artificial neural network was trained using the backpropagation paradigm on a microcomputer.
• The network was trained on numeral ultrasound data from 52 patient cases (17 malignant, 30 cysts, 5 other).
• The trained prototype was tested on 47 cases not included in the training dataset.

Main Results:

• The artificial neural network correctly identified renal cell carcinoma from renal cysts and other conditions.
• The prototype achieved zero diagnostic errors on the trained cases.
• The trained ANN prototype performed without error on 47 independent test cases.

Conclusions:

• The prototype artificial neural network demonstrates high accuracy in diagnosing renal cell carcinoma from ultrasound data.
• ANNs show significant potential as a tool for providing reliable second opinions in clinical diagnosis.
• Further validation through extended studies with more cases is recommended to confirm the findings.