Table 3.

Literature review of AI/radiomics studies in the veterinary imaging applications, with reported accuracies and conclusions. CNN: convolutional neural networks. N/A: not available.

Reference	Topic	Scale	Species	AI/Radiomic Algorithms	Accuracy	Conclusion
Basran et al., 2021 [63]	Lesion detection: equine proximal sesamoid bone micro-CT	Clinical N = 8 cases and 8 controls	Equine	Radiomics	N/A	Radiomics analysis of μCT images of equine proximal sesamoid bones was able to identify image feature differences in image features in cases and controls
Becker et al., 2018 [64]	Lesion detection: murine hepatic MRIs	Pre-clinical N = 8 cases and 2 controls.	Murine	Radiomics	N/A	Texture features may quantitatively detect intrahepatic tumor growth not yet visible to the human eye
Boissady et al., 2020 [65]	Lesion detection: canine and feline thoracic radiographic lesions	Clinical N = 6584 cases	Canine and feline	Machine learning -CNN	N/A	The described network can aid detection of lesions but not provide a diagnosis; potential to be used as tool to aid general practitioners
McEvoy and Amigo, 2013 [62]	Lesion detection: canine pelvic radiograph classification	Clinical N = 60 cases	Canine	Machine learning -CNN	N/A	Demonstrated feasibility to classify images, dependent on availability of training data
Yoon et al., 2018 [66]	Lesion detection: canine thoracic radiographic lesions	Clinical N = 3122 cases	Canine	Machine learning -CNN -BOF	CNN: 92.9–96.9% BOF: 79.6–96.9%	Both CNN and BOF capable of distinguishing abnormal thoracic radiographs, CNN showed higher accuracy and sensitivity than BOF
Banzato et al., 2018 [67]	Lesion characterization: MRI differentiation of canine meningiomas vs. gliomas	Clinical N = 80 cases	Canine	Machine learning -CNN	94% on post-contrast T1 images, 91% on pre-contrast T1-images, 90% on T2 images	CNN can reliably distinguish between different meningiomas and gliomas on MR images
D’Souza et al., 2019 [68]	Lesion characterization: assessment of B-mode US for murine hepatic fibrosis	Pre-clinical N = 22 cases and 4 controls.	Murine	Radiomics	N/A	Quantitative analysis of computer-extracted B-mode ultrasound features can be used to characterize hepatic fibrosis in mice
Kim et al., 2019 [69]	Lesion characterization: canine corneal ulcer image classification	Clinical N = 281 cases	Canine	Machine learning -CNN	Most models > 90% for superficial and deep corneal ulcers; ResNet and VGGNet > 90% for normal corneas, superficial and deep corneal ulcers	CNN multiple image classification models can be used to effectively determine corneal ulcer severity in dogs
Wanamaker et al., 2021 [70]	Lesion characterization: MRI differentiation of canine glial cell neoplasia vs. noninfectious inflammatory meningoencephalitis	Clinical N = 119 cases	Canine	Radiomics	Random forest classifier accuracy was 76% to differentiate glioma vs. noninfectious inflammatory meningoencephalitis	Texture analysis using random forest algorithm to classify inflammatory and neoplastic lesions approached previously reported radiologist accuracy, however performed poorly for differentiating tumor grades and types