Neural networks

Compiled by

Dean Giustini, UBC Biomed librarian, dean.giustini@ubc.ca

Updated

5 December 2025 | Part of Knowledge Synthesis (KS) & Artificial Intelligence (AI) Search Wiki 2025 & A to Z Listing

Introduction

A neural network (artificial neural network or neural net, abbreviated ANN or NN) is a computational model powering a lot of modern AI. Neural networks, also convolutional neural networks (CNNs), enable machines to learn from data, similar to the human brain. A NN consists of interconnected nodes, or "neurons," organized in layers—input, hidden, and output. Each neuron processes input, applies weights, biases, and activation functions, and passes the result forward. Through training, typically via backpropagation and gradient descent, neural networks adjust these parameters to minimize errors, excelling in tasks such as image recognition, natural language processing, and predictive analytics.

Deep neural networks, with multiple hidden layers, handle complex patterns but require significant computational resources and data. Architectures like CNN specialize in spatial data, such as images, while recurrent neural networks (RNNs) excel in sequential data, such as time series or text. Transformers, a newer architecture, dominate natural language tasks due to their attention mechanisms, which prioritize relevant data points. Incorporating searching and indexing into neural networks enhances efficiency, especially in large-scale applications.

The US National Library of Medicine's automated indexing system, called the MTIX, uses neural networks to perform AI-powered indexing. The NLM deployed the Medical Text Indexer (neXt generation) MTIX in 2024, and performs automated indexing within a 24 hour window.

Presentation *

Note: This presentation was selected by a librarian for its introduction as well as its concision and accuracy in conveying information about the topic.

Medical text indexer (MTIX) and automated indexing

The Medical Text Indexer (MTI), developed by the National Library of Medicine (NLM), uses neural networks for the MTIX (Medical Text Indexer-NeXt Generation). It uses machine learning to assign Medical Subject Headings (MeSH) to articles, improving indexing speed and scalability. Trained on millions of MEDLINE citations from 2007–2022, the MTIX analyzes titles, abstracts, and journal metadata to recommend relevant MeSH terms with high recall (e.g., >94% for disease detection) and precision (e.g., 87% for disease categories). The MTI supports semi-automated and fully automated indexing, reducing human indexer workload while maintaining quality. For medical texts, MTIX processes full-text articles when available, improving term coverage over title-and-abstract-based methods. Filtering techniques, such as ranking scores and excluding lengthy documents, further boost accuracy.

Neural networks in MTIX enable rapid, precise indexing, critical for scaling up to the growing volume of biomedical literature - in 2024, 1.5 million papers. While human curation remains in place in MEDLINE for quality control, MTIX’s automation project and use of AI supports applications such as the publicly-available MeSH on Demand tool, aiding researchers in metadata identification.

References

Disclaimer

Note: Please use your critical reading skills while reading entries. No warranties, implied or actual, are granted for any health or medical search or AI information obtained while using these pages. Check with your librarian for more contextual, accurate information.