Chondral Bone Lesion Characterization
| Purpose | Provide a probability of a chondral bone lesion having benign vs malignant potential |
| Tag(s) |
|
| Panel | Musculoskeletal |
| Define-AI ID | 180050002 |
| Originator | Martin Torriani |
| Panel Chair | Jay Patti |
| Panel Reviewers | Musculoskeletal Panel |
| License | |
| Status | Published |
Clinical Implementation
Value Proposition
Bone tumors containing cartilage (enchondromas, low-grade chondrosarcomas) can have similar radiographic appearances. This struggle in differentiating an enchondroma (benign lesion) from a low-grade chondrosarcoma (malignant lesion) is a long-held diagnostic dilemma for which no satisfactory diagnostic support is available. A dedicated algorithm trained to differentiate between these two entities may detect features that are not readily noticeable to human readers. This diagnostic dilemma can be particularly burdensome to nonspecialists (radiologists, pathologists, and orthopedists), who would benefit from an algorithm providing a probability distribution for benign versus malignant lesions. AI meeting this use case would help to reduce the false negative rate, patient risk, and the medical legal risk for the radiologists.Narrative(s)
An adult patient presents with a lytic lesion suggestive of enchondroma on an x-ray (incidental or not). The algorithm evaluates the image and provides a probability of enchondroma, chondrosarcoma, or unknown. The radiologist is informed of this information at the time of interpretation.Workflow Description
An image is obtained from the modality and sent to PACS and the AI engine. The image is analyzed by the engine. The system segments the lesion and analyzes in comparison with a model trained from known cases of each class. A message is sent to PACS from the engine with the classification information. If the location of the lesion is also identified, location information can also be sent to PACS to highlight the region the engine identified.
Consideration for future work would be to provide a combined probability score derived from the assessment of multiple modalities. Therefore, this system may perform best if x-ray and CT data are provided, rather than x-ray alone. In this scenario, the system would need to be trained for multiple modalities.
Considerations for Dataset Development
Procedures(s): {X-ray, Lower/Upper Extremity; X-ray Pelvis; CT; MRI}
View(s): Various obliquities of patient on the images
Sex at Birth: {Male, Female}
Age: [18,90]
Data Source: {Referral Centers; Community Clinics}
Location of tumor: {Hand, Humerus, Femur, Tibia, Ribs, Pelvis}
Differential Diagnoses: {bone infarcts, fibrous lesions, bone cysts, degenerative cysts}
Anatomic Location: {Hand, Humerus, Femur, Tibia, Ribs, Pelvis}
Diagnosis (Radiology/Pathology): Preference for referral centers. Use data that tracks patients over time to verify enchondroma. Ideally, data comes from bone pathologists with experience in sarcoma imaging.
Other: {Fracture, Pain}
Technical Specifications
Inputs
DICOM Study
| Procedure | X-ray, Lower/Upper Extremity; X-ray Pelvis |
| Data Type | DICOM |
| Modality | XRAY |
Primary Outputs
Chondrosarcoma Probability Score
| RadElement ID | |
| Definition | Calculates the probability of a given lesion to be low-grade chondrosarcoma. Values closer to 1 indicate chondrosarcoma. Values closer to 0 indicate enchondroma. |
| Data Type | Numeric |
| Value Set | [0,1] |
| Units | N/A |
Related Datasets
No known related public datasets at this time, please alert us if you know of any.