Lung Cancer |

Performance of a Next Generation Computer-Aided Detection Algorithm for Detection of Missed Lung Cancers on Chest Radiography FREE TO VIEW

Calen Frolkis, BA; Dr. Robert Gilkeson, BA
Author and Funding Information

Case Western Reserve University-University Hospitals, Cleveland, OH

Chest. 2014;146(4_MeetingAbstracts):596A. doi:10.1378/chest.1967264
Text Size: A A A
Published online


SESSION TITLE: Lung Cancer Posters II

SESSION TYPE: Original Investigation Poster

PRESENTED ON: Wednesday, October 29, 2014 at 01:30 PM - 02:30 PM

PURPOSE: This restrospective study sought to evaluate a next generation CAD algorithm in the detetion of previously overlooked lung cancer on chest radiographs.

METHODS: IRB approval was granted for this retrospective study. Patient consent was waived given the retrospective data collected. Review of 606 patients presented at our Institutional Thoracic Tumor Board was performed. Clinical reports from cross-sectional imaging, pathology, and prior chest Xrays were analyzed. Of these, 41 patients lung cancer was initially overlooked on CXR. A board certified chest radiologist with 15yrs experience confirmed the location of disease. A subtlety rating from 1-10 (1: very subtle, 10: very obvious) was assigned to each CXR. Anatomical distribution of nodules was recorded. All 41 images were then analyzed by a next-generation CAD algorithm (OnGuard 5.2 Riverain Medical). The time lapse between index CXR and diagnosis was recorded.

RESULTS: There were 41 missed diagnoses. The age range was 44-91, mean age: 68.75. There were 23 women, and 18 men. Subtlety ranged from 1-9 (mean 2.39; mode 2). Missed lesions were likely to be in the Right Upper Lobe and apex (41.4%), and Left Upper Lobe (19.5%). Time lapse between the reported 'normal' index CXR and diagnosis ranged from 1 day to 3070 days, (8 years). The average time lapse was 577.28 days. The most common pathology was Adenocarcinoma (24, 58.5%), Squamous cell carcinoma (14, 34.1%) and metastatic lesions (3 cases, 7%). New generation CAD detected 28 of 41 lesions (sensitivity: 67%). Most lesions seen by CAD were found in the upper lobes and apices (13, 31.7% RT and 8, 19.5% LT). There was a FP average of .75 per image. Most FPs were in the upper lobes, often overlying bony structures, and pleural surfaces. Subtlety ranged 1-9 (average 2.67; mode 2). CAD had 13 false negatives.

CONCLUSIONS: This retrospective review shows that the new generation 5.2 CAD system is superior in detection of lung nodules with fewer FPs than previous generations.

CLINICAL IMPLICATIONS: Prior literature has demonstrated that early lung cancer is often overlooked on chest radiography. While early efforts demonstrated improved detection rates with the use of CAD, high FP rates has limited clinical adoption. This study sought to evaluate diagnostic performance in the detection of lung cancer with a new generation CAD software, and prove a much lower FP rate per image than previously reported.

DISCLOSURE: Dr. Robert Gilkeson: Consultant fee, speaker bureau, advisory committee, etc.: Riverain The following authors have nothing to disclose: Calen Frolkis

No Product/Research Disclosure Information




Citing articles are presented as examples only. In non-demo SCM6 implementation, integration with CrossRef’s "Cited By" API will populate this tab (http://www.crossref.org/citedby.html).

Some tools below are only available to our subscribers or users with an online account.

Related Content

Customize your page view by dragging & repositioning the boxes below.

Find Similar Articles
CHEST Journal Articles
PubMed Articles
  • CHEST Journal
    Print ISSN: 0012-3692
    Online ISSN: 1931-3543