Endoscopic ultrasound is an important tool for the assessment of patients with pancreatic cystic lesions. These lesions are common findings, and may convey an increased risk of pancreatic malignancy. Nevertheless, this risk is almost exclusive of a group of pancreatic cystic lesions – mucinous cystic lesions. The differentiation between mucinous and non-mucinous lesions bases on endoscopic ultrasound is limited by limited interobserver agreement.
Our group has developed a pioneer proof-of-concept study to assess the performance of a deep learning algorithm for the automatic identification and differentiation of pancreatic mucinous cystic lesions versus normal pancreatic parenchyma or non-mucinous lesions. This pilot algorithm achieved a sensitivity of 98%, a specificity of 99%, and an overall accuracy of 99%. Our group believes that this work is a fundamental cornerstone for further development of these algorithms which will contribute to mitigate the current limitations of endoscopic ultrasound in this setting.
The paper can be found here.
In Press // Diagnostics 2022, 12(9), 2041 // August 2022