Lung Cancer
Conditions
Keywords
Lung Cancer, Chest Radiograph, Computer-aided Detection, Image processing
Brief summary
This is a clinical trial using retrospective data of two different software devices developed by Riverain Medical Group: Softview and OnGuard 5.0. The two studies will be run concurrently. Riverain Medical Group's computer systems are designed to assist radiologists in their identification of lung cancer on chest radiographs. The current machine received FDA Pre-Market Approval. This is to test two new software approaches.
Detailed description
In 2000, data was presented to the FDA to demonstrate that a new system for computer analysis could assist radiologists in the detection of small lung cancers on chest radiographs. Radiologists using the system showed a statistically significant improvement in lung cancer detection rate when they used the system, compared to their interpretation of chest radiographs when they did not use the computer system. This study, along with other supporting data, resulted in the FDA giving Pre-Market Approval for the system. The system has undergone several improvements in software and hardware, and it is now intended to test two different software systems to determine whether radiologists using the systems can improve their detection of lung cancer on chest radiographs. One of these systems processes the chest radiograph to decrease the emphasis given to the shadow of the ribs and thereby enhances the ability of radiologists to detect disease in the lungs. The second system performs a series of evaluations on chest radiographs and, based on a complex system of analysis, points to locations on the chest radiograph that contain solitary pulmonary nodules having the characteristics of primary lung cancer or solitary metastases of cancer to the lungs. This will be a test of radiologists to determine the degree of improvement, if any, that results when they interpret chest radiographs that may or may not have cancer, first interpreted without the computer and, second, with the images output by the software.
Interventions
BEHAVIORALSoftware
This is an observer performance study. Radiologists will interpret chest radiographs without and then with the Riverain software, both SoftView (TM) OnGuard (TM) CADe Software with be tested
Sponsors
Riverain Technologies
Georgetown University
Eligibility
Sex/Gender
ALL
Healthy volunteers
Yes
Inclusion criteria
* US American Board of Radiology Certified Radiologists in active clinical practice
Exclusion criteria
* Specialists in pulmonary or chest or cardio-pulmonary radiology Prior membership on expert panels for this study who prepared cases Current or recent colleagues or trainees (within 10 years) of the Principal Investigator
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Improvement in Cancer Detection as Measured by Localized Receiver Operating Characteristic) LROC Changes Under the LROC Curve. | Three days of experiment over 3-5 months, varied by participant | Standard methods for LROC methodology and statistical analysis were used. We are testing two different types of software using different cases, but the same radiologists to control for radiologist differences. LROC is Localized Receiver Operating Characteristic. LROC measures the trade-offs between sensitivity and specificity as radiologists use different levels of suspicion of disease. This analysis is for the software that decreases the visibility of the ribs and clavicles while preserving (and potentially enhancing) the visibility of the lungs and lung diseases. In this case, the level of suspicion recorded was for the radiologist's concern that a finding did or did not represent cancer. Please note that the FDA approved indications for use is to detected nodules that may represent cancer, but in our study scoring for a true finding was based on whether or not the nodule did represent cancer. A larger number, if statistically significant, indicates that that method is better. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Sensitivity and Specificity Using SoftView Software | Three days of experiment over 3-5 months, varied by participant | Sensitivity and specificity were calculated using the radiologists' responses of recommendations for follow-up with CT or biopsy. Truth was whether or not the nodule identified was found to be cancer. Sensitivity is the percentage of correct identification of a positive case (a case with cancer). Specificity is the percentage of negative cases (those without cancer) that were correctly identified as not having cancer. The mean values of 15 radiologists are used. |
Other
| Measure | Time frame | Description |
|---|---|---|
| Difference in the Area Under the LROC Curve Comparing OnGuard 1.0 and OnGuard 5.1 | 5 months | This reports the comparison of the detection of lung nodules that were proven to represent lung cancers. It compares the results of two versions of computer-aided detection software: OnGuard 1.0 from 2001 and OnGuard 5.1 from 2009. The results represent the responses of radiologists when they use one or the other types of software. To compare radiologists' results with the two types of software, the measurement analyzed was the difference in the areas under the localized receiver operating characteristic curve (LROC). The results from the 15 participating radiologists were averaged (mean value). The area under the LROC curve is a measure of the trade-offs between sensitivity and 1-specificity that occurs as the level of certainty of a positive finding changes. It is normally reported as a decimal without units. In this study dsign, a lower number indicates that the new method (OnGuard 5.1), if statistically significant, if better. |
Countries
United States
Participant flow
Recruitment details
Recruitment between April 9, 2009 and June 15, 2009. Recruited by email and phone calls to individuals meeting entry criteria specified in the protocl.
Pre-assignment details
All recruited individuals met entry criteria. None were excluded or dropped from the study. Training in the use of studied device (software) occurred immediately prior to the experiment. Training took 55 to 80 minutes, depending on the individuals speed. Each individual served as his/her own control, so all participants were in both groups
Participants by arm
| Arm | Count |
|---|---|
| Radiologists Radiologists who have certification by the American Board of Radiology
Riverain OnGuard CAD Software : This is an observer performance study. Radiologists will interpret chest radiographs without and then with the Riverain software, both SoftView (TM) OnGuard (TM) CADe Software with be tested | 15 |
| Total | 15 |
Baseline characteristics
| Characteristic | Radiologists |
|---|---|
| Age, Customized number of participants | 15 participants |
| Region of Enrollment United States | 15 participants |
| Sex: Female, Male Female | 5 Participants |
| Sex: Female, Male Male | 10 Participants |
Adverse events
| Event type | EG000 affected / at risk |
|---|---|
| deaths Total, all-cause mortality | — / — |
| other Total, other adverse events | 0 / 15 |
| serious Total, serious adverse events | 0 / 15 |
Outcome results
Primary
Improvement in Cancer Detection as Measured by Localized Receiver Operating Characteristic) LROC Changes Under the LROC Curve.
Standard methods for LROC methodology and statistical analysis were used. We are testing two different types of software using different cases, but the same radiologists to control for radiologist differences. LROC is Localized Receiver Operating Characteristic. LROC measures the trade-offs between sensitivity and specificity as radiologists use different levels of suspicion of disease. This analysis is for the software that decreases the visibility of the ribs and clavicles while preserving (and potentially enhancing) the visibility of the lungs and lung diseases. In this case, the level of suspicion recorded was for the radiologist's concern that a finding did or did not represent cancer. Please note that the FDA approved indications for use is to detected nodules that may represent cancer, but in our study scoring for a true finding was based on whether or not the nodule did represent cancer. A larger number, if statistically significant, indicates that that method is better.
Time frame: Three days of experiment over 3-5 months, varied by participant
Population: 122 subjects had cancer that potentially could be detected on their chest radiograph. Power analysis showed that sample size of 351 patients, in a 2:1 ratio of nodule absent to present patients was selected to provide 80% power to detect a difference in areas under the curve of 0.10 or greater.
| Arm | Measure | Value (MEAN) |
|---|---|---|
| Radiologists Control for SoftView | Improvement in Cancer Detection as Measured by Localized Receiver Operating Characteristic) LROC Changes Under the LROC Curve. | 0.460 unitless |
| Radiologists Using Softview Software | Improvement in Cancer Detection as Measured by Localized Receiver Operating Characteristic) LROC Changes Under the LROC Curve. | 0.558 unitless |
Comparison: The sample size of 351 patients, in a 2:1 ratio of nodule absent to present patients was selected to provide 80% power to detect a difference in areas under the curve of 0.10 or greater. This sample size was calculated using PASS software (Hintze, 2008). Settings: • α = 0.025. • one-sided test • AUCA = 0.80 and AUCSV = 0.90.• 2:1 ratio of patients with nodules to those without • areas calculated for the entire curves • correlation 0.3\* • discrete data. • standard deviation ratios of 1\*.p-value: 0.05Bootstraping
Secondary
Sensitivity and Specificity Using SoftView Software
Sensitivity and specificity were calculated using the radiologists' responses of recommendations for follow-up with CT or biopsy. Truth was whether or not the nodule identified was found to be cancer. Sensitivity is the percentage of correct identification of a positive case (a case with cancer). Specificity is the percentage of negative cases (those without cancer) that were correctly identified as not having cancer. The mean values of 15 radiologists are used.
Time frame: Three days of experiment over 3-5 months, varied by participant
| Arm | Measure | Group | Value (MEAN) |
|---|---|---|---|
| Radiologists Control for SoftView | Sensitivity and Specificity Using SoftView Software | Sensitivity | 49.5 percentage of cases |
| Radiologists Control for SoftView | Sensitivity and Specificity Using SoftView Software | Specificity | 96.1 percentage of cases |
| Radiologists Using Softview Software | Sensitivity and Specificity Using SoftView Software | Sensitivity | 66.3 percentage of cases |
| Radiologists Using Softview Software | Sensitivity and Specificity Using SoftView Software | Specificity | 91.8 percentage of cases |
Other Pre-specified
Difference in the Area Under the LROC Curve Comparing OnGuard 1.0 and OnGuard 5.1
This reports the comparison of the detection of lung nodules that were proven to represent lung cancers. It compares the results of two versions of computer-aided detection software: OnGuard 1.0 from 2001 and OnGuard 5.1 from 2009. The results represent the responses of radiologists when they use one or the other types of software. To compare radiologists' results with the two types of software, the measurement analyzed was the difference in the areas under the localized receiver operating characteristic curve (LROC). The results from the 15 participating radiologists were averaged (mean value). The area under the LROC curve is a measure of the trade-offs between sensitivity and 1-specificity that occurs as the level of certainty of a positive finding changes. It is normally reported as a decimal without units. In this study dsign, a lower number indicates that the new method (OnGuard 5.1), if statistically significant, if better.
Time frame: 5 months
Population: 81 of the 263 radiographs contained a non-calcified nodule that had been diagnosed as lung cancer. Power calculation showed 246 patients, in a 2:1 ratio of nodule absent to present would provide 80% power to detect a difference in areas under the curve of 0.10 or greater.
| Arm | Measure | Value (MEAN) |
|---|---|---|
| Radiologists Control for SoftView | Difference in the Area Under the LROC Curve Comparing OnGuard 1.0 and OnGuard 5.1 | -0.043 unitless |