Page 60 - Read Online

Tovar et al. Art Int Surg 2023;3:14-26 https://dx.doi.org/10.20517/ais.2022.38 Page 18

Table 2. Artificial Intelligence definitions
AI ML ANN DL CNN
A family of Subset of AI that learns Subset of ML designed to Subset of ANNs that contain A class of ANN that uses
computational to convert input data mimic biological neural multiple neural layers between mathematical convolution -
methods designed into a desired output networks. ANNs include the input and output layers. application of a pattern filter to
to mimic human based on analysis of computational neurons These networks may contain small fields within the data in a
[28]
intelligence and training data . composed of an input billions of parameters (e.g, GPT- manner similar to the human
decision-making Common ML methods layer, at least one hidden 3 at 175B [30] ) that form complex visual cortex - to interpret
include support vector layer, and an output representations of patterns from imaging or audio data [32] . There
machines, decision trees, layer [29] training datasets [31] is overlap between CNN and DL
and Bayesian networks networks
AI: Artificial Intelligence; ANN: artificial neural network; CNN: convolutional neural network; DL: deep learning neural network; ML: machine
learning.

Table 3. Summary of recent AI models utilized in PDAC research
Authors Year Model description AI algorithm Results
Risk prediction:
Boursi et al. [38] 2017 Early-onset diabetes, health data Multivariable logistic regression AUC, 0.82
[39]
Boursi et al. 2022 Impaired fasting glucose Multivariable model AUC, 0.71
diagnosis, health data
[40]
Muhammad et al. 2019 Health data, 18 features ANN Training test AUC, 0.86
Test set AUC, 0.85
[41]
Qureshi et al. 2022 Risk prediction using radiomics Bayes classifier 86% accuracy
[42]
Chen et al. 2020 Pancreatic ductal dilation Multi-state model AUC, 0.825-0.833
Early detection:
[43]
Mukherjee et al. 2022 Early detection using radiomics ML AUC, 0.98
Permuth et al. [44] 2016 IPMN classification, CT and Logistic regression AUC, 0.92
miRNA
Polk et al. [45] 2020 IPMN classification, CT Multivariate model AUC, 0.93
[46]
Tobaly et al. 2020 IPMN classification, CT Multivariate model AUC, 0.84
Kuwahara et al. [47] 2019 IPMN classification, EUS DL AUC, 0.98
[48]
Hanania et al. 2016 IPMN classification, CT Logistic regression AUC, 0.96
Momeni-Boroujeni 2017 FNA biopsy malignancy MNN Stratification of atypical cases as benign or
[49]
et al. malignant, 77% accuracy
[50]
Chen et al. 2022 Detection of tumors (< 2cm) CNN Internal test, AUC 0.96
using radiomics Test set, AUC 0.95
[51]
Zhang et al. 2022 Detection of cancer clusters, EUS- DCNN Internal test, AUC 0.958
FNA External test, AUC 0.948-0.976
[52]
Kartal et al. 2022 Fecal microbiome Classifier AUC, 0.94
Zaid et al. [53] 2020 Classification of tumors as high Logistic regression-based binary AUC, 0.84
delta or low delta classification
ANN: Artificial neural network; CNN: convolutional neural network; CT: computed tomography; DCNN: deep convoluted neural network; DL: deep
learning; EUS-FNA: endoscopy ultrasonography-fine-needle aspiration; ML: machine learning; MNN: multilayer perceptron neural network; PDAC:
pancreatic ductal adenocarcinoma carcinoma.

Using data from nearly 800,000 patients, an ANN was developed by incorporating 18 personal health
features from datasets. These variables included data that are ubiquitous in health records, such as the
presence of diabetes, race, and family history. The model stratified patients as low-, medium-, and high-risk
[40]
and performed with an AUC of 0.86 .
Image-based risk models
The term “radiomics” refers to a family of image analysis techniques that convert image data into sets of
quantitative feature measurements that represent key features like brightness, shape, and texture. The