Related Experiment Video
Updated: May 4, 2026
Quantitation of Protein Expression and Co-localization Using Multiplexed Immuno-histochemical Staining and Multispectral Imaging
Published on: April 8, 2016
Multimodal AI Combining Clinical and Imaging Inputs Improves Prostate Cancer Detection
Christian Roest1, Derya Yakar, Dorjan Ivan Rener Sitar
1From the Department of Radiology, Medical Imaging Center, University Medical Center Groningen, Groningen, the Netherlands (C.R., D.Y., D.I.R.S., S.J.F., T.C.K.); Department of Radiology, Netherlands Cancer Center Antoni van Leeuwenhoek, Amsterdam, the Netherlands (D.Y.); Department of Radiology, Radboud University Medical Center, Nijmegen, the Netherlands (J.S.B., H.H.); and Department of Radiology, Martini Ziekenhuis Groningen, Groningen, the Netherlands (D.B.R.).
This study shows that combining deep learning (DL) with clinical data improves prostate cancer detection on MRI. Multimodal artificial intelligence (AI) integrating DL suspicion levels and clinical parameters offers superior diagnostic accuracy for clinically significant prostate cancer (csPCa).
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Deep learning (DL) models for prostate cancer detection on MRI often neglect key clinical factors.
- Integrating clinical parameters like PSA, prostate volume, and age could enhance DL diagnostic accuracy for clinically significant prostate cancer (csPCa).
Purpose of the Study:
- To explore the integration of clinical parameters with MRI-based DL for improved csPCa detection.
- To evaluate multimodal artificial intelligence (AI) classifiers combining DL features and clinical data.
Main Methods:
- Retrospective analysis of 932 biparametric prostate MRI exams.
- Development of six multimodal AI classifiers using early and late fusion of DL suspicion levels, clinical parameters, and lesion volumes.
- Internal and external validation of AI models against radiologist assessments using PI-RADS.
Main Results:
- The multimodal AI integrating DL suspicion levels and clinical features via early fusion demonstrated superior external performance (AUC 0.77).
- Early fusion outperformed late fusion (AUC 0.77 vs 0.73 externally).
- Multimodal AI performance showed no significant difference compared to radiologist assessments (external AUC 0.77 vs 0.75).
Conclusions:
- Multimodal AI combining DL suspicion levels and clinical parameters surpasses MRI-only AI and clinical parameters alone for csPCa detection.
- Early information fusion enhances AI robustness in a multicenter setting.
- Incorporating lesion volumes did not improve diagnostic efficacy.
