Extranodal Extension
| Purpose | Detect/delineate lymph node involvement and extranodal extension on cross-sectional images |
| Tag(s) |
|
| Panel | Oncology |
| Define-AI ID | 18070001 |
| Originator | Oncology Panel |
| Panel Chair | Reid F. Thompson |
| Panel Reviewers | Oncology Panel |
| License | |
| Status | Published |
| RadElement Set | RDES30 |
Clinical Implementation
Value Proposition
Canonically, ECE is determined after time of surgical excision, often connoting a substantially worse prognosis at that time. This use case would be of most relevance for diagnoses where surgery occurs after a period of neoadjuvant therapy, and could enable treatment intensification prior to the finding of ECE at time of surgery. Moreover, a performant algorithm could potentially identify ECE for diagnoses that do not usually proceed to surgery, potentially enabling better treatment stratification in this population. Automated ECE classification and identification could also enable improved radiotherapy targeting of nodal basins, as well as treatment optimization for post-operative imaging-detected nodal disease. Specific examples of these scenarios include but are not limited to:
- Head and neck cancers
- Prostate cancer
- Anal and colorectal cancers
- Cervical and endometrial cancers
Although not proven, this algorithm or a semi-automated approach could improve cancer outcomes and decrease morbidity.
Narrative(s)
60 year old male smoker undergoes CT of the head and neck for his newly diagnosed cancer of the oral cavity. Algorithm evaluates image and identifies all visible lymph nodes, classifies each of them as radiographically normal, involved by cancer, or indeterminate, and further identifies presence and location of any radiographically visible extranodal extension. If a radiologist is not present at the time of imaging, an alert is provided to the ordering physician. Algorithm results will be accessible as a DICOM-RT structure set.Workflow Description
Image obtained from modality and sent to PACS and the AI engine. Image analyzed by engine. System detects and defines lymph node(s) and assigns probability of malignancy as well as presence of extranodal extension. An alert message is sent to PACS from the engine with the information, identification, and graphic highlighting segmented and labeled normal and abnormal lymph nodes as a DICOM-RT structure set object.Considerations for Dataset Development
Technical Variance
Modality: {CT (helical, cone beam)}
Contrast: {agent, dose, route, protocol}
Scanner: {manufacturer, age, model, tabletop}
Setup devices: {aquaplast mask, breast board, etc.}
Positioning: {neck flexed/extended, arms down/up, legs frog-legged, etc.}
Artifacts: {dental or orthopedic hardware, patient motion, pixel loss}
Acquisition protocol: {scanning parameters (e.g. slice thickness), pulse sequence, etc.}
Clinical Variance
Anatomical site: {ensure dataset includes supraclavicular, axillary, iliac, inguinal, and other areas of lymphadenopathy in addition to cervical and retropharyngeal lymphadenopathy}
Tumor type: {SCC, adenocarcinoma, salivary gland histologies, melanoma, other}
Viral status: {HPV subtypes, EBV, HIV}
Lymph node size: {numerous examples of sub-centimeter disease extending all the way to bulky lymphadenopathy}
Habitus: {height/weight/BMI, algorithm should be agnostic to cachexia, obesity, etc.}
Age: {algorithm should account for cases in juvenile/pediatric as well as very elderly contexts}
Competing diagnoses: {acute infection (e.g. viral), chronic infection (e.g. TB), autoimmune (e.g. SLE), lymphoma}
Confounders: {prior XRT, prior surgery or SLNB, prior trauma, birth defects}
Demographics: {sex, ethnicity}
Technical Specifications
Inputs
DICOM Study
| Procedure | CT |
| Data Type | DICOM |
| Modality | CT |
| Anatomic Focus | Any |
| Scenario | Cancer Diagnosis |
Primary Outputs
Lymph Node Identification
| RadElement ID | RDE207 |
| Definition | Detect and delineate visible lymph nodes |
| Data Type | DICOM-RT structure set |
| Value Set | 3D structure coordinates |
| Units | N/A |
| Multiplicity | 1 (single structure set returned with all detected lymph nodes) |
Lymph Node Classification
| RadElement ID | |
| Definition | Classify individual lymph nodes as radiographically normal, involved by cancer, or indeterminate |
| Data Type | Categorical |
| Value Set | Unknown radiographically normal involved by cancer |
| Units | N/A |
| Multiplicity | [0,∞] (repeated for each detected lymph node) |
Future Development Ideas
Related Applications
- Segment and describe lymph nodes with one or more of the RadElement descriptors (RDES24 )
- Integration into cancer staging paradigms and workflows
- Monitoring for disease recurrence
- Computer-assisted or fully automated radiation treatment planning
Challenges
- Segmentation may provide significant difficulty depending on factors such as fibrosis, any post treatment changes, fat content, anatomical location, and image artifacts, among others.
- Availability of pathologic confirmation of disease and localization of ECE may be particularly challenging at the dataset construction level
Related Datasets
The Cancer Imaging Archive (TCIA)
Know of more related datasets? Please let us know