top of page

AI-based diagnosis of skeletal metastases in whole-body MRI

By: Jakub Ceranka, Joris Wuts, Ophélye Chiabai, Frédéric Lecouvet and Jef Vandemeulebroucke

AI-based diagnosis of skeletal metastases in whole-body MRI

The integration of artificial intelligence (AI) in healthcare has recently witnessed a remarkable surge in popularity. AI algorithms, used in medical imaging to automate the detection and quantification of the infiltration of various cancers, not only significantly decrease the workload of a radiologist, but also often improve the diagnostic accuracy. Computer-aided diagnosis (CAD) systems are particularly important in diseases requiring a large amount of data to be reviewed (e.g. patient whole-body imaging), which makes their manual quantification labor demanding and prone to error.

Metastatic bone disease - a secondary cancer originating in the prostate or breasts and spreading to the skeleton in the form of bone metastases; often results in multiple cancer infiltration sites with scattered distribution of metastases of complex shapes. Early detection of such bone involvement is important for disease staging, therapeutic decisions, and evaluation of the patient treatment response.

A coronal slice of whole-body MR modalities used in the CAD system. Bone metastases are marked with red arrows. Lesions represent different intensity profiles compared to the healthy bone, dependent on the used imaging modality

Whole-body magnetic resonance imaging (MRI) has been successfully used for monitoring of metastatic bone disease, allowing for comprehensive and holistic evaluation of metastatic bone disease and treatment response assessment. The major challenges of radiological reading of whole-body MRI come from the amount of data to be reviewed. Additionally, whole-body MRIs are often corrupted with multiple spatial and intensity distortions, which further degrade the performance of image reading and image processing algorithms.

A schematic graphical representation of the proposed computer-aided diagnosis system.

The close collaboration between the Department of Electronics and Informatics (ETRO) of the VUB and the Institut de Recherche Expérimentale et Clinique (IREC) of the Cliniques universitaires Saint Luc in Brussels, resulted in the development of the first fully automated computer-aided diagnosis system for the detection and quantification of metastatic bone disease using whole-body MRI. The system consists of an extensive image preprocessing pipeline aiming at enhancing the image quality of acquired MRI images, followed by a deep learning framework for detection and segmentation of metastatic bone disease.


Because of the 3D nature and high resolutions of medical images, deep learning pipelines in the medical field require more resources compared to classical computer vision approaches. The A100 GPUs available at the HPC allowed us to perform the experiments.

‘Training single segmentation model already required considerable computational resources. Moreover, cross-validation strategies were needed for valid comparisons between multiple proposed methods. The pool of GPU nodes available at the VSC allowed for parallel training of the different models, facilitating the computing needs of this research.’

An example segmentation result of multiple bone metastases in three different anatomical sites (columns). The top row represents manual ground-truth segmentation prepared by an expert radiologist. The bottom row is the output of the proposed detection system with a corresponding Dice coefficient.

The system outperformed state-of-the-art methodologies, achieving a detection sensitivity of 63% with a mean of 6.44 false positives per image, and an average lesion Dice coefficient of 0.53. A provided ablation study performed to investigate the relative importance of image preprocessing shows that the introduction of the region of interest mask and spatial registration have a significant impact on detection and segmentation performance in whole-body MRI. The proposed computer-aided diagnosis system allows for automatic quantification of disease infiltration and could provide a valuable tool during radiological examination of whole-body MRI.



Read the full publication in the ‘Computer Methods and Programs in Biomedicine' here.


bottom of page