Artificial Intelligence, Endobronchial Ultrasound, Lung Cancer
Conditions
Brief summary
To evaluate the usefulness of Deep neural network (DNN) in the evaluation of mediastinal and hilar lymph nodes with Endobronchial ultrasound (EBUS). The study will explore the feasibility of DNN to identify lymph nodes and blood vessel examined with EBUS.
Detailed description
Multi-center prospective feasibility study. The DNN model will be trained on ultrasound images with annotation to identifies lymph nodes and blood vessels examined with EBUS. The ability of the DNN to segment lymph nodes and vessels based on postoperative processing and static EBUS images will be evaluated in the first part of the study. In the second part of the study Real-time use of DNN in EBUS procedure will be evaluated.
Interventions
Machine learning algorithm run on EBUS images for real-time labelling of mediastinal lymph nodes and lymph node level
Sponsors
Helse Nord-Trøndelag HF
SINTEF Health Research
Norwegian University of Science and Technology
Study design
Observational model
CASE_ONLY
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
* Subjects referred to thoracic department in any of the participating hospitals with undiagnosed enlarged mediastinal and hilar lymph nodes. * Subjects have to be ≥ 18 years of age
Exclusion criteria
* Pregnancy * Any patient that the Investigator feels is not appropriate for this study for any reason.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Capability | 8 months | To explore if Deep neural network (DNN) has capability to segment lymph nodes and blood vessels from EBUS images |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Sensitivity | 2 months | True positive rate. Correctly detected lymph nodes/blood vessel over total lymph nodes/blood vessel. Measured per pixel in the EBUS images |
| Specificity | 2 months | Specificity = (True Negative)/(True Negative + False Positive). Measured per pixel in the EBUS images. |
| Precision | 2 months | The precision the DNN has for detecting lymph nodes and blood vessels. Measured both per voxel in the EBUS images and per annotated structure (a structure is counted as detected if at least 50% of its annotated pixels are identified by the DNN). |
| Run-time | 2 months | Is the run-time sufficiently low for real-time analysis during EBUS? |
| Adverse events | 48 hours | Procedure related adverse events or unexpected incidents registered |
| Dice similarity coefficient | 2 months | Measures the similarity between two sets of data: Annotated by pulmonologist vs DNN. |
Countries
Norway
Contacts
Primary ContactØyvind Ervik, MD
Backup ContactHanne Sorger, MD,PhD
Outcome results
None listed