Artificial neural networks: opening the black box.

Area of Science:

• Computational biology
• Medical informatics
• Artificial intelligence in medicine

Background:

• Artificial neural networks (ANNs) are established computational methods with strong theoretical foundations.
• ANNs demonstrate significant potential for application across various disciplines, particularly in medicine.
• Emerging applications highlight the need for understanding ANNs in medical contexts.

Purpose of the Study:

• To provide a clear understanding of biologically inspired computing paradigms for medical applications.
• To review the historical development and mathematical underpinnings of ANNs, focusing on multilayered perceptrons.
• To discuss best practices in training, validation, testing, and performance assessment of ANNs in medicine.

Main Methods:

• Discussion of historical development and basic operational mathematics of multilayered perceptrons.
• Description of training, validation, and testing techniques for ANNs.
• Explanation of performance and reliability measurements, including bootstrap methods and receiver operating characteristic (ROC) methodology.

Main Results:

• ANNs facilitate the extraction of novel medical information from raw data.
• Computer models built with ANNs aid medical decision-making and distribute expertise.
• ANNs enable personalized patient outcome predictions, shifting from traditional 'bin-model' approaches.

Conclusions:

• ANNs offer a paradigm shift towards individualized patient management and prediction.
• Applications in prostate cancer detection, heart disease risk, and medication dosing show promise.
• Overcoming obstacles like data limitations and establishing multidisciplinary teams is crucial for widespread ANN adoption in healthcare.