Related Experiment Video
Updated: Nov 2, 2025

Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
Deep Learning for Classification of Bone Lesions on Routine MRI.
Feyisope R Eweje1, Bingting Bao2, Jing Wu2
1Department of Radiology, Children's Hospital of Philadelphia, Philadelphia, PA, 19104, USA; Perelman School of Medicine at University of Pennsylvania, Philadelphia, PA, 19104, USA.
A new deep learning model can differentiate benign and malignant bone lesions using MRI scans and patient data, performing comparably to expert radiologists. This AI tool could improve diagnostic accuracy and reduce unnecessary procedures.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Distinguishing benign from malignant bone lesions is challenging for radiologists due to overlapping imaging features.
- Routine magnetic resonance imaging (MRI) and patient demographics are key data sources for bone lesion analysis.
Purpose of the Study:
- To develop a deep learning algorithm for differentiating benign and malignant bone lesions.
- To assess the algorithm's performance against human expert classification.
Main Methods:
- A dataset of 1,060 bone lesions from five institutions was used for model development, internal validation, and external testing.
- EfficientNet-B0 architecture was employed for image-based models, combined with a logistic regression model using patient age, sex, and lesion location.
- A voting ensemble model was created and its performance compared to radiology experts.
Main Results:
- The deep learning ensemble model achieved an accuracy of 0.76, sensitivity of 0.79, and specificity of 0.75, with a ROC AUC of 0.82.
- Performance was comparable to expert radiologists (accuracy 0.73, sensitivity 0.81, specificity 0.66).
- External testing demonstrated a ROC AUC of 0.79 for the model.
Conclusions:
- Deep learning models can effectively distinguish between benign and malignant bone lesions, matching expert performance.
- These AI-driven tools have the potential to enhance diagnostic accuracy and streamline patient referrals, reducing unnecessary biopsies and specialized center visits.
Related Concept Videos
Classification of Bones
Long and Short Bones
The appendicular skeleton, particularly the upper and lower limbs, is primarily made of long and short bones. The...
Magnetic Resonance Imaging
Gross Anatomy of Bone
The diaphysis is the tubular shaft that runs between the proximal and distal ends of the bone. The walls of the diaphysis are composed of dense and hard compact bone made of numerous osteons — the functional unit of the compact bone. The hollow region in the diaphysis is called the medullary cavity, which harbors the bone marrow. In infants and children, this marrow cavity is filled with red marrow, whereas in...
