Page 78 - Read Online

Page 381 Novotny et al. Art Int Surg 2024;4:376-86 https://dx.doi.org/10.20517/ais.2024.52

representative training sets, the automatically learned features often outperform the manually designed
features due to their high selectivity and invariance. This autonomous learning process analyzes an
enormous number of cases, outperforming human experts. If the training data are sufficiently diverse, DL
[39]
can cover a wide range of case variations . The application of DL can be used in breast reconstruction
surgery, focusing on how DL models can help predict the best flap type and forecast aesthetic outcomes.
The structure of AI [Figure 1] shows the steps from a simple DL network to a complete AI. AI uses and
combines the processes of DL, ANN, and ML networks. This enables comprehensive data analysis and has
the great potential to deliver the most accurate and appropriate solutions.

BIG DATA AND DATA PROTECTION
Big data is the foundation of AI, providing the raw material with which AI works. AI draws its knowledge
from huge databases . It is not just the amount of data or its size that defines big data, but rather the
[41]
enormous complexity and diversity of the information. The information in these databases can be
structured, semi-structured, or unstructured. In the healthcare sector, modernization has resulted in an
almost infinite amount of patient data that do not all conform to the same categories or standards due to
local differences. These databases are, therefore, ripe for big data .
[42]
One challenge presented by big data is ensuring the privacy and proper handling and protection of patient-
related data. The best possible protection of patient data can only be achieved through cooperation between
patients, medical staff, administrators, the legal system, and the government. General resources, rules, and
conditions for data protection and data handling must be established and enforced . Probably the most
[43]
important requirement for big data is the complete anonymization of patient data. In this context, the term
“anonymized” is used to describe data that cannot be identified as belonging to a particular person, either
on its own or in combination with other data. This is achieved by taking into account all potential means
that could reasonably be used to identify an individual. The process of anonymization forms the basis for
the protection of personal data [44,45] .

AI IN OTHER MEDICAL FIELDS
One of the primary applications of AI is in radiology. In a 2016 study, an artificial learning model was used
to analyze mammograms, addressing common issues such as overdiagnosis, lack of time, and inaccuracy in
current diagnostic methods . In another study from Heidelberg, Germany, an AI network trained with
[46]
12,378 open-source dermatoscopic images for melanoma image classification outperformed 136 of 157
dermatologists regardless of their experience levels . These results underscore the substantial learning
[47]
capacity of AI and its ability to leverage massive databases effectively.

ETHICAL AND LEGAL ASPECTS
One of the biggest challenges at present is the uncertainty of the underlying data on which these systems are
trained. The sources from which the data are drawn cannot be verified for timeliness, reliability, accuracy,
and validity due to their sheer volume. In addition, these sources are not exclusively related to plastic
surgery or the required specialty. Due to the lack of control over the quality and relevance of this training
data, there is a risk of producing information that is not based on scientific evidence . Generally, AI relies
[48]
on existing data to make suggestions. As a result, it is essential to ensure the quality of the data and to select
appropriate data sets for specific patient groups. In addition, the data collected must be permitted for use
and anonymized to fully protect individual privacy before being included in the dataset. A data use
agreement is required . Patient data have to be protected against misuse, with clear limitations established
[5]
on the scope of AI applications.