einstein (São Paulo). 17/Dec/2020;19:eED6037.

Unpacking the black box

Thiago Gonçalves dos Santos

Martins

, Paulo

Schor

DOI: 10.31744/einstein_journal/2021ED6037

Today there are large databases (Big Data) of electronic records and digital images that allow pattern recognition in large volumes of information in a short period of time, contributing to the creation of a personalized Medicine with the aid of artificial intelligence.^(,)

The algorithms used in medical diagnosis software are formed by a deep neural network, consisting of several layers that resemble the biological processes of the human cortex. Each neuron corresponds to a specific stimulus for a region within an image, similar to the way the brain neuron would respond to visual stimuli, activating a specific region of visual space. The neural network learns automatically by conferring values to these filters, even though specific parameters such as number of filters, filter size, and network architecture still need to be defined before the training phase.

[…]