Implementation of deep learning in liver pathology optimizes diagnosis of benign lesions and adenocarcinoma metastasis [data]

• Verfasst von: Kriegsmann, Mark [VerfasserIn]
• Verfasst von: Kriegsmann, Katharina [VerfasserIn]
• Verfasst von: Steinbuß, Georg [VerfasserIn]
• Verfasst von: Zgorzelski, Christiane [VerfasserIn]
• Verfasst von: Albrecht, Thomas [VerfasserIn]
• Verfasst von: Heinrich, Stefan [VerfasserIn]
• Verfasst von: Farkas, Stefan [VerfasserIn]
• Verfasst von: Roth, Wilfried [VerfasserIn]
• Verfasst von: Hausen, Anne [VerfasserIn]
• Verfasst von: Gaida, Matthias [VerfasserIn]
• Titel: Implementation of deep learning in liver pathology optimizes diagnosis of benign lesions and adenocarcinoma metastasis [data]
• Verf.angabe: Mark Kriegsmann, Katharina Kriegsmann, Georg Steinbuss, Christiane Zgorzelski, Thomas Albrecht, Stefan Heinrich, Stefan Farkas, Wilfried Roth, Anne Hausen, Matthias M. Gaida
• Verlagsort: Heidelberg
• Verlag: Universität
• E-Jahr: 2023
• Jahr: 2023-07-07
• Umfang: 1 Online-Ressource (11 Files)
• Fussnoten: Gesehen am 20.07.2023
• DOI: doi:10.11588/data/YAZWJW
• Datenträger: Online-Ressource
• Dokumenttyp: Forschungsdaten
• Dokumenttyp: Datenbank
• Sprache: eng
• Bibliogr. Hinweis: Forschungsdaten zu: Kriegsmann, Mark, 1987 - : Implementation of deep learning in liver pathology optimizes diagnosis of benign lesions and adenocarcinoma metastasis
• Sach-SW: Health and Life Sciences
• Sach-SW: Medicine
• K10plus-PPN: 1853174386
• K10plus-PPN:
  Universitätsbibliothek
• Lokale URL UB: Zum Volltext

Abstract

Differentiation of neoplastic and non-neoplastic liver lesions using routine histological tissue sections can be challenging. Correct classification is paramount to forecast prognosis and to select the correct therapy. Deep learning algorithms have recently been suggested to support objective and consistent assessment of digital histopathological images. In thisstudy, annotation of 7 different classes, namely non-neoplastic bile ducts, benign biliary lesions and liver metastases from colorectal and pancreatic adenocarcinoma, was performed, resulting in a total of 204.159 image patches. The patient cohort was split into three datasets and an EfficientNetV2 and ResNetRS deep learning algorithm to classify the respective categories was trained, optimized, and ultimately tested. Model performance was evaluated on validation and test data using confusion matrices. In summary, a hereinafter proposed automated classification to identify benign and malignant liver lesions by deep learning methods was described, which performed with high diagnostic accuracy. Furthermore, a huge curated liver dataset was provided.