RT Journal Article SR Electronic T1 Clinical decision support for axillary lymph node staging in newly diagnosed breast cancer patients based on 18F-FDG PET/MRI and machine-learning JF Journal of Nuclear Medicine JO J Nucl Med FD Society of Nuclear Medicine SP jnumed.122.264138 DO 10.2967/jnumed.122.264138 A1 Janna Morawitz A1 Benjamin Sigl A1 Christian Rubbert A1 Nils-Martin Bruckmann A1 Frederic Dietzel A1 Lena J. Häberle A1 Saskia Ting A1 Svjetlana Mohrmann A1 Eugen Ruckhäberle A1 Ann-Kathrin Bittner A1 Oliver Hoffmann A1 Pascal Baltzer A1 Panagiotis Kapetas A1 Thomas Helbich A1 Paola Clauser A1 Wolfgang Peter Fendler A1 Christoph Rischpler A1 Ken Herrmann A1 Benedikt M. Schaarschmidt A1 Andreas Stang A1 Lale Umutlu A1 Gerald Antoch A1 Julian Caspers A1 Julian Kirchner YR 2022 UL http://jnm.snmjournals.org/content/early/2022/09/22/jnumed.122.264138.abstract AB Background: In addition to its high prognostic value, the involvement of axillary lymph nodes in breast cancer patients also plays an important role in therapy planning. Therefore, an imaging modality that can determine nodal status with high accuracy in primary breast cancer patients is desirable. Purpose: To investigate if machine-learning prediction models based on simple assessable imaging features in MRI (magnetic resonance imaging) or PET (positron emission tomography)/MRI are able to determine nodal status in newly diagnosed breast cancer patients with comparable performance as experienced radiologists, if such models can be adjusted to achieve low rates of false negatives such that invasive procedures could potentially be omitted, and if a clinical framework for decision-support based on simple imaging features can be derived from these models. Methods: 303 participants from three centres prospectively underwent dedicated whole-body 18F-FDG (18F-fluorodeoxyglucose) PET/MRI between August 2017 and September 2020. Imaging datasets were evaluated regarding axillary lymph node metastases based on morphologic and metabolic features. Predictive models were developed for MRI and PET/MRI separately using random forest classifiers on data of two centers and were tested on data of the third center. Results: The diagnostic accuracy for MRI features was 87.5% both for radiologists and for machine learning algorithm. For PET/MRI the diagnostic accuracy was 89.3% for the radiologists and 91.2% for the machine learning algorithm with no significant differences in diagnostic performance of radiologists and the machine learning algorithm in MRI (P = 0.671) and PET/MRI (P = 0.683). Most important lymph node feature was tracer uptake, followed by lymph node size. With an adjusted threshold, a sensitivity of 96.2% was achieved by the random forest classifier, whereas specificity, positive predictive value, negative predictive value and accuracy were 68.2%, 78.1%, 93.8% and 83.3%. A decision tree based on three simple imaging features could be established for MRI and PET/MRI. Conclusion: Applying a high sensitivity threshold to the random forest results could potentially avoid invasive procedures such as sentinel lymph node biopsy in 68.2% of the patients.