Related Experiment Video
Updated: Sep 15, 2025
Label-Free Identification of Lymphocyte Subtypes Using Three-Dimensional Quantitative Phase Imaging and Machine Learning
Published on: November 19, 2018
Deep Learning Applications in Lymphoma Imaging
Vera Sorin1, Israel Cohen2, Ruth Lekach3
1Department of Radiology, Mayo Clinic, Rochester, Minnesota, USA.
Background:
Lymphomas are a diverse group of disorders characterized by clonal proliferation of lymphocytes. While definitive diagnosis relies on histopathology, immunohistochemical, molecular and genomic analyses, imaging modalities including positron emission tomography/computed tomography (PET/CT), computed tomography (CT), and magnetic resonance imaging (MRI) are essential in diagnostic processes and management. Imaging aids in detecting suitable biopsy sites, assessing disease extent, evaluating treatment response, and detecting recurrence. However, accurate diagnosis and staging remain challenging due to tumor heterogeneity, inter-observer variability, and technical imaging issues.
Summary:
Artificial intelligence (AI), particularly deep learning (DL) models, is transforming lymphoma imaging by enabling automated detection, segmentation, and classification across PET/CT, CT, and MRI modalities. Key applications include automated metabolic response assessment and total metabolic tumor volume quantification in PET/CT, lymph node segmentation and classification in CT, and improved detection of central nervous system involvement in MRI. Despite promising results, significant challenges limit widespread clinical adoption, including variability in imaging protocols affecting model generalizability, reliance on small retrospective datasets, lack of model interpretability, and difficulties integrating AI tools into existing clinical workflows.
Key Messages:
(1) DL applications can automate detection, segmentation, and classification in lymphoma imaging, improving diagnostic accuracy and reducing inter-observer variability across PET/CT, CT, and MRI modalities. (2) Challenges in DL adoption include validating model performance across diverse imaging protocols, addressing data biases, and ensuring generalizability to real-world clinical settings. (3) Integrating AI into clinical workflows requires careful validation to ensure safety, consistency, and alignment with existing diagnostic and treatment standards.
