Improving Gastric Cancer Outcome Prediction Using Single Time-Point Artificial Neural Network Models.

Area of Science:

• Oncology
• Biostatistics
• Artificial Intelligence in Medicine

Background:

• Predicting cancer outcomes using prognostic variables is crucial in oncology.
• Traditional survival analysis models often rely on restrictive assumptions (e.g., proportional hazards).
• Artificial neural networks (ANNs) offer a powerful nonlinear approach for complex data analysis in medicine.

Purpose of the Study:

• To develop and evaluate an ANN model for predicting gastric cancer patient survivability.
• To address the challenge of censored data in survival analysis for gastric cancer.
• To provide accurate, nonlinear predictions of cancer outcome at multiple time points.

Main Methods:

• Development of five distinct single time-point Artificial Neural Network (ANN) models.
• Models were designed to predict patient outcomes at 1, 2, 3, 4, and 5 years post-diagnosis.
• Inclusion and handling of censored data were integral to the model development.

Main Results:

• The ANN models demonstrated consistently high performance in predicting gastric cancer survival probabilities.
• Accuracy and area under the receiver operating characteristic curve (AUC) were high across all tested time points.
• The models effectively managed censored data, enhancing prediction reliability.

Conclusions:

• Artificial neural network models provide a robust and accurate method for predicting gastric cancer survivability.
• This nonlinear approach overcomes limitations of traditional survival analysis, especially with censored data.
• The developed ANN models show significant potential for clinical application in cancer outcome prediction.