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

Leah Kathleen Pomerantz, Mauricio Solano, Eric Kalosa-Kenyon

Background: The page is an original research article published in Veterinary Radiology & Ultrasound, a peer-reviewed journal. The article evaluates the performance of a commercial artificial intelligence (AI) software for detecting pulmonary nodules and masses in canine thoracic radiographs.

Study: The study is a retrospective, diagnostic accuracy design that uses confirmed cases of pulmonary nodules and masses based on CT, cytology, or histopathology as the ground truth. The study also uses a control group of normal cases without pulmonary nodules. The study aims to test the hypothesis that the AI software has high positive and negative predictive values and accuracy for pulmonary nodule and mass detection.

Methods: The study uses 56 cases with confirmed pulmonary nodules and masses and 32 control cases without nodules. The cases are submitted to the AI software (Vetology AI®), which uses a convolutional neural network (CNN) to classify images as positive or negative for a nodular pattern. The software’s output is compared with the ground truth and the sensitivity, specificity, predictive values, accuracy, balanced accuracy, and F1-score are calculated. The study also analyzes the effect of lesion size, location, shape, opacity, and superimposition on the software’s performance.

Results: The study finds that the AI software correctly detects pulmonary nodules and masses in 31 of the 56 confirmed cases and correctly classifies 30 of the 32 control cases. The software’s accuracy is 69.3%, balanced accuracy 74.6%, F1-score 0.7, sensitivity 55.4%, and specificity 93.8%. The study also finds that the software performs better on lesions that are larger, solitary, in the left caudal or right cranial lung lobes, or diffuse/multifocal, and worse on lesions that are smaller, in the right caudal, accessory, left cranial, or right middle lung lobes, or cavitated.

Limitations: The study acknowledges several limitations, such as the low number of cases, the lack of distinction between nodules and masses, the inclusion of cases with confounding factors or incomplete radiographic studies, the use of cases from a single institution, and the lack of transparency in the AI software’s training and testing data.

Conclusions: The study concludes that the AI software has moderate performance for pulmonary nodule and mass detection in canine thoracic radiographs, and that further validation and improvement of the software are needed. The study also highlights the importance of transparency in AI software development and the potential role of AI tools in veterinary diagnostic imaging.

Disclaimer: The summary generated in this email was created by an AI large language model. Therefore errors may occur. Reading the article is the best way to understand the scholarly work. The figure presented here remains the property of the publisher or author and subject to the applicable copyright agreement. It is reproduced here as an educational work. If you have any questions or concerns about the work presented here, reply to this email.