High Risk Cancer
Conditions
Brief summary
This is a Multinational, Multicenter, retrospective study for the evaluation of the standalone efficacy and safety of an Artificial Intelligence/Machine Learning (AI/ML) technology-based end-to-end Computer assisted Detection/Computer Assisted Diagnosis (CADe/CADx) Software as a Medical Device (SaMD) developed to detect, localize and characterize malignant, and suspicious for lung cancer nodules on Low Dose Computed Tomography (LDCT) scans taken as part of a Lung Cancer Screening (LCS) program. LDCT Digital Imaging and Communications in Medicine (DICOM) images of patients who underwent lung cancer screening were selected and included into the study. Selected scans will then be analyzed by the CADe/CADx SaMD and compared to radiologist generated reference standards including lesions localization and lesion cancer diagnosis. Figures of merit at patient level and lesion level detection and diagnostic efficacy will be calculated as well as sub-class analysis to ensure algorithm performance generalizability.
Interventions
DEVICEMedian LCS
End-to-end processing of chest LDCT DICOM images by an AI/ML tech-based SaMD to detect, localize, and characterize (assign a malignancy score) each detected pulmonary nodule. The output of the device is a DICOM File (Median LCS result report) summarizing results per patient.
Sponsors
Median Technologies
Study design
Observational model
COHORT
Time perspective
RETROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
50 Years to 80 Years
Healthy volunteers
Yes
Inclusion criteria
* ≥50-80 Years of age; * Current or ex-smoker (\>=20 pack years); * Patient screened and surveilled for lung cancer screening following lung cancer screening guidelines (equivalent to United States Preventive Services Task Force (USPSTF) 2021 Criteria); * Received LDCT due to inclusion in high-risk category for lung cancer.
Exclusion criteria
* Prior lung resection; * Pacemaker or other indwelling metallic medical devices in the thorax that interfere with CT acquisition; * Patients/images used during AI model development; * Patients with only hilar and/or mediastinal cancer(s); * Patients with only ground glass cancer(s); * Patients with nodules, solid or part-solid \>30mm (masses); * Patients that are not accompanied with the required clinical information; * Patients with imaging with any of the following: missing slices, slice thickness \>3mm; * Partial cover of the lung.
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| AUROC (Area under ROC curve) at patient level | 12 months | AUROC that measures Median LCS performance at patient level is strictly superior to 0.8. Support for Primary Endpoint: Derived from the patient level AUROC at the product fixed operating point : Sensitivity, Specificity, PPV, NPV. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Specificity > 70% when Sensitivity=70% | 12 months | — |
| AUC of LROC > 0.75 | 12 months | In contrast to the receiver operating characteristic (ROC) assessment paradigm, localization ROC (LROC) analysis provides a means to jointly assess the accuracy of localization and detection in an observational study. |
| Detection sensitivity>0.8 with average FP rate per scan<1 | 12 months | — |
| ICC>0.8 for average diameter | 12 months | Intraclass Correlation Coefficient (ICC), is a descriptive statistic that can be used when quantitative measurements are made on units that are organized into groups. It describes how strongly units in the same group resemble each other. |
| Sensitivity > 70% when Specificity=70% | 12 months | — |
| ICC>0.8 for short axis diameter | 12 months | — |
| ICC>0.75 for Volume | 12 months | — |
| DICE Coefficient >0.7 | 12 months | — |
| ICC>0.8 for long axis diameter | 12 months | — |
Countries
Spain, United States
Outcome results
None listed