Artificial intelligence 101 for veterinary diagnostic imaging

Abstract

The prevalence and pervasiveness of artificial intelligence (AI) with medical images in veterinary and human medicine is rapidly increasing. This article provides essential definitions of AI with medical images with a focus on veterinary radiology. Machine learning methods common in medical image analysis are compared, and a detailed description of convolutional neural networks commonly used in deep learning classification and regression models is provided. A brief introduction to natural language processing (NLP) and its utility in machine learning is also provided. NLP can economize the creation of "truth‐data" needed when training AI systems for both diagnostic radiology and radiation oncology applications. The goal of this publication is to provide veterinarians, veterinary radiologists, and radiation oncologists the necessary background needed to understand and comprehend AI‐focused research projects and publications.

Performance of a commercially available artificial intelligence software for the detection of confirmed pulmonary nodules and masses in canine thoracic radiography

“…Although the limitations of this specific training dataset may be important in the context of this study, it may be considered that regardless of current shortcomings, the accuracy of AI models may increase with increasingly larger datasets that include a wider range of disease presentations. 17 Veterinary models, including Vetology, often utilize a few hundred to a few thousand training images, as compared with human counterparts that often use orders of magnitude more cases. 14,24,25 Thus, it may be hypothesized that the accuracy of the Vetology model may improve by adding a breadth of cases to the training dataset, though further testing following this retraining would be necessary to test this hypothesis.…”

“…16 The method of analysis for labeling images (e.g., "pulmonary nodule positive" or "normal") is called "supervised learning", whereby the AI model learns to correlate patterns in images with their associated labels. 17 In a binary classification setting, for each image, the output of the computer vision model is a prediction score from 0 to 1. As there is uncertainty in model prediction (represented by values between 0 and 1), an image is classified as "positive" when its prediction value is higher than a fixed "decision threshold."…”

“…A ROC curve depicts the effect of changing the decision threshold on the ratio of false positives to false negative results and can be used to evaluate the diagnostic ability of a binary classifier system using a variety of potential decision thresholds. 12,18 These metrics, as evaluated internally by the AI company, are not readily provided to the general commercial user, 17 but are instead often held as proprietary intellectual information. To the authors' knowledge, no commercial veterinary AI tool has had transparency in their data collection or model development, as has been suggested as one of the hallmarks of "Good Machine Learning Practices" by the Food and Drug Administration (FDA): Guiding Principle 9: "Users Are Provided Clear, Essential Information.…”

Introduction to the Veterinary Radiology & Ultrasound Special Issue on Artificial Intelligence

“…2 This will provide a baseline understanding and direct you to prior publications you may have missed. From there Hespel et al 3 will provide a tutorial on the fundamental concepts in AI, Basran and Porter 4 will explore the same in radiomics and Joslyn and Alexander will provide you with the steps needed to evaluate AI products as you begin to consider their application in practice. 5 Sections 2 and 3 of this issue contain a number of reviews and commentaries highlighting the potential applications of AI and inviting you to consider some of the challenges and opportunities we may face as we implement these technologies in veterinary practice.…”

“…This will provide a baseline understanding and direct you to prior publications you may have missed. From there Hespel et al 3 . will provide a tutorial on the fundamental concepts in AI, Basran and Porter 4 will explore the same in radiomics and Joslyn and Alexander will provide you with the steps needed to evaluate AI products as you begin to consider their application in practice 5 .…”