Back to the list
Congress: ECR24
Poster Number: C-21633
Type: EPOS Radiologist (scientific)
DOI: 10.26044/ecr2024/C-21633
Authorblock: C. Salvatore1, M. Interlenghi1, E. Schiavon1, A. Lad1, D. Fazzini1, M. Alì1, S. Papa1, F. Sardanelli2, I. Castiglioni3; 1Milan/IT, 2San Donato Milanese, Milan/IT, 3Milano/IT
Disclosures:
Christian Salvatore: Shareholder: DeepTrace Technologies S.R.L., Milan, Italy CEO: DeepTrace Technologies S.R.L., Milan, Italy
Matteo Interlenghi: Shareholder: DeepTrace Technologies S.R.L., Milan, Italy Employee: DeepTrace Technologies S.R.L., Milan, Italy
Elia Schiavon: Employee: DeepTrace Technologies S.R.L., Milan, Italy
Akash Lad: Nothing to disclose
Deborah Fazzini: Nothing to disclose
Marco Alì: Other: Bracco Imaging
Sergio Papa: Nothing to disclose
Francesco Sardanelli: Nothing to disclose
Isabella Castiglioni: Shareholder: DeepTrace Technologies S.R.L., Milan, Italy
Keywords: Breast, Mammography, Computer Applications-General, Cancer
Purpose The purpose of this work was to assess the real-world performance of a deep learning algorithm integrated with a PACS and a mammography unit-vendor neutral system, for the automatic classification of breast density into 4 categories, according to the ACR BI-RADS.
Read more Methods and materials We conducted a multicentric real-world study by including patients who performed digital mammography (DM) or synthetic 2D-views from digital breast tomosynthesis (s2D-views) at 3 centers of Centro Diagnostico Italiano (Milan, Italy). DICOM images of these patients were sent automatically through secure DICOM-communication protocol into the central PACS (Picture Archiving and Communication System) of the center.The Trace4BDensity™ deep learning tool (DeepTrace Technologies, Italy) was installed on a server of the center as an on-premise service and integrated with a PACS research...
Read more Results A total of 567 patients was processed by the Trace4BDensity™ tool. For each patient, a report was automatically provided to the PACS research node and made available at a patient level.It must be noted that the classification results were inaccurate in presence of image artifacts associated to a breast incorrect positioning, which occurred in 25 of 567 patients (4.4%).Overall, the breast density classification proposed by the model was dense in 55% of the patients, and nondense in 45% of the...
Read more Conclusion Our results show that a real-world use of a PACS-integrated deep-learning system to classify breast density on digital mammography or on synthetic 2D-views from digital breast tomosynthesis is feasible and reliable. Image artifacts associated to patient's breast incorrect positioning may lead to misclassification errors, with low occurrence rates.
Read more References Magni, V., Interlenghi, M., Cozzi, A., Alì, M., Salvatore, C., Azzena, A. A., ... & Sardanelli, F. (2022). Development and validation of an AI-driven mammographic breast density classification tool based on radiologist consensus. Radiology: Artificial Intelligence, 4(2), e210199.
Read more