Page 168 - Read Online

Endo et al. Art Int Surg 2024;4:59-67 https://dx.doi.org/10.20517/ais.2024.09 Page 61

of data, ML algorithms facilitate the identification of data patterns and enhance data-based predictions with
incrementally improved performance optimization over time. There are various types of ML approaches,
[21]
such as supervised learning and unsupervised learning . In addition, ML can utilize several different types
of algorithms that can recognize patterns within a set of data including artificial neural networks, decision
tree algorithms [i.e., random forest (RF), gradient-boosting] [22,23] , and instance-based algorithms (i.e.,
support vector machine, k-nearest neighbor) [24,25] . Deep learning (DL) is a subset of ML that involves neural
networks with multiple layers (i.e., deep neural networks) and has been demonstrated to be particularly
effective in handling complex tasks. DL has gained prominence within the broader field of AI due to its
ability to process data in a hierarchical fashion, thereby allowing for more sophisticated and nuanced
[26]
decision making . DL techniques, particularly convolutional neural networks (CNN), have been applied to
radiomics, which is a field of medical imaging that involves extracting a large number of quantitative
features from medical images . NLP encompasses the interaction between computers and human
[27]
language, including tasks such as speech recognition, language translation, sentiment analysis, and language
understanding. NLP has given rise to applications such as chatbots, language translation services, and voice-
activated assistants [28-30] . NLP enables the handling of vast “unstructured” text-based data, and can enhance
the ability to make clinical diagnoses and risk stratification of patients [20,31] .

Visual AI
Visual AI involves the utilization of artificial intelligence techniques to analyze and comprehend visual
information. Visual AI involves empowering machines to interpret, understand, and make decisions based
on visual data (i.e., intraoperative images and videos). Visual AI applications leverage technologies such as
computer vision, DL, and various image processing techniques to extract meaningful insights from visual
content. The significance of visual AI has gained considerable attention, especially in its potential
applications in the field of surgery . For instance, the automation of surgical phase recognition, utilizing
[32]
AI and computer vision algorithms for image interpretation through DL, has been a subject of recent
interest. This application has demonstrated its effectiveness in enhancing surgeon performance in both
basic procedures like inguinal hernia repair, as well as complex surgeries such as robot-assisted minimally
invasive esophagectomy and hepatectomy [33,34] . In the field of surgical procedures, particularly in liver
surgery, in which the understanding of complex anatomy is crucial to improve perioperative outcomes, the
implementation of AI as an operative aid is likely to become a hot topic in the future.

Various applications of AI to HB cancer patients
Table 1 summarizes several representative studies related to the application of AI to HB cancer patients.

Hepatocellular carcinoma
Treatment strategies for hepatocellular carcinoma (HCC) are diverse and usually based on factors such as
tumor morphology, biology, patient performance status, and background liver condition . The decision-
[35]
making process becomes even more complex in the retreatment of patients with recurrent lesions post
initial liver resection, as many different multidisciplinary approaches need to be considered . To this end,
[36]
some researchers have sought to establish optimal treatment ML algorithms to inform the care of these
patients [37,38] . For example, Famularo et al. examined an Italian registry of patients with HCC and an external
Japanese cohort to develop and validate a prediction model related to survival after recurrence (SAR) .
[37]
This model, based on a standard Cox model that incorporated all second-order interactions of treatment
with features selected by the Least Absolute Shrinkage and Selection Operator (LASSO) , included
[39]
treatment-related variables and time to recurrence. The resulting ML-based model demonstrated high
predictive ability, as shown by the area under the curve (AUC) values (3-year SAR: AUC 0.805; 5-year SAR:
AUC 0.785). The authors concluded that an ML-based model could assist in allocating treatment for
patients with recurrent HCC . Another study by our own research group utilized an international, multi-
[37]