A Formal validation of an Entropy-based Artificial Intelligence for Ultrasound Data in Breast Tumors

Abstract

Abstract Background: Research on artificial intelligence-assisted breast diagnosis is mainly based on static images or dynamic videos. The acquired images or videos may come from ultrasound probes of different frequencies. It is not clear how frequency-induced image variations affect the diagnosis of artificial intelligence models. Purpose: To explore the impact of using ultrasound images of variable frequencies on the diagnostic efficacy of artificial intelligence in breast ultrasound screening. Materials and Methods: Video and entropy-based, using a feature entropy breast network compared the diagnostic performance and average two-dimensional image entropy of the L14-L9 linear array probe and L13-L7 linear array probe. Results: In testing set 1, the diagnostic efficiency of the L9 dataset is better than L14; In testing set 2, the diagnostic efficiency of the L13 dataset is better than L7; the value of L9, L13 dataset is greater than L14, L7dataset in the average two-dimensional image entropy, respectively. Conclusion: Ultrasound images obtained with a certain degree of lower frequency probes have a higher average two-dimensional image entropy, which is beneficial for the diagnosis of artificial intelligence models. The higher the average two-dimensional image entropy of the dataset, the superior its diagnostic performance.