Combining ultrasound radiomics, complete blood count, and serum biochemical biomarkers for diagnosing intestinal disorders in cats using machine learning

Parminder S. Basran, Natalya Shcherban, Marnin Forman, Jasmine Chang, Sophie Nelissen, Benjamin K. Recchia, Ian R. Porter

Background: Imaging alone cannot distinguish between inflammatory and neoplastic diseases of the feline bowel. Combining radiomics with other methods may prove valuable in differentiating and predicting bowel disorders.

Study: The study was a retrospective analytical observational cohort study that included 149 cats from three institutions, who had biopsy-confirmed diagnoses of small cell epitheliotropic lymphoma, inflammatory bowel disease, no pathology, or other conditions. The study used various combinations of features extracted from transverse sections of small intestine ultrasound images, complete blood count, and serum biochemical profile data to train different machine learning models and evaluate their performance.

Methods: The study used two feature selection methods (wrapper and decision tree) to identify the top 20 features from five datasets of different combinations of ultrasound radiomics, complete blood count, and serum biochemical profile data. The study also used six machine learning algorithms (support vector machine, K-nearest neighbor, random forest, RUSBoost random forest, neural network, and Naïve Bayes) to train and test models for four classification schemes: (1) normal versus abnormal; (2) warranting or not warranting a biopsy; (3) lymphoma, inflammatory bowel disease, healthy, or other conditions; and (4) lymphoma, inflammatory bowel disease, or other conditions. The study used 10-fold cross-validation and accuracy as the main performance metric.

Results: The study found that the models based on ultrasound radiomics data alone or in combination with complete blood count and serum biochemical profile data achieved relatively high accuracies for the first two classification schemes (0.866 and 0.759, respectively), but lower accuracies for the third and fourth classification schemes (0.504 and 0.531, respectively). The study also found that the feature selection methods did not substantially differ in their predictive performance, and that the support vector machine, K-nearest neighbor, and random forest algorithms performed consistently well across the datasets and classification schemes.

Limitations: The study acknowledged some limitations, such as the potential bias in the healthy cohort data, the batch effects from the multi-institutional data, the manual segmentation of the ultrasound images, the use of a single transverse section for radiomics extraction, and the lack of autosegmentation and additional imaging features.

Conclusions: The study concluded that machine learning models can be developed to classify feline intestinal abdominal abnormalities by combining ultrasound radiomics, complete blood count, and serum biochemical profile data, and that these models could offer benefits in clinical scenarios where the cost–risk–benefit analysis of obtaining a gastrointestinal biopsy is unclear. The study also suggested some directions for future research, such as improving the data quality and quantity, exploring other radiomic features and machine learning methods, and conducting prospective clinical trials.

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.