Chest
Volume 128, Issue 3, September 2005, Pages 1517-1523
Journal home page for Chest

Clinical Investigations
Computer-Aided Diagnosis as a Second Reader: Spectrum of Findings in CT Studies of the Chest Interpreted as Normal

https://doi.org/10.1378/chest.128.3.1517Get rights and content

Study objectives

To assess the performance of an automated computer-aided detection (CAD) system as a second reader on chest CT studies interpreted as normal at routine clinical interpretation.

Design

Chest CT studies were processed using a prototype CAD system for automated detection of lung lesions. Three experienced radiologists analyzed each CAD finding and confirmed or dismissed the marked image features as lung lesions. Noncalcified, focal lung lesions were classified according to size as being of high (≥ 10 mm), intermediate (5 to 9 mm), or low (≤ 4 mm) significance.

Setting

Two subspecialized academic tertiary referral centers in the United States and Germany.

Patients

Chest CT studies were performed in 100 patients, with results initially reported as normal at clinical double reading. Indications for chest CT were suspected pulmonary embolism (PE) [n = 33], lung cancer screening in a high-risk population (n = 28), or follow-up for a cancer history (n = 39).

Interventions

Reevaluation of all chest CT studies for focal lung lesions with the CAD system as a second reader.

Measurements

Prevalence and spectrum of lung lesions missed at routine clinical interpretation but found by the CAD system.

Results

In 33% (33 of 100 patients), CAD detected significant lung lesions that were not previously reported. Fifty-three significant lesions were detected (mean, 1.6 lesions per case), of which 5 lesions (9.4%) were of high significance, 21 lesions (39.6%) were of intermediate significance, and 27 lesions (50.9%) were of low significance. In the PE group, the lung cancer screening group, and the group with a cancer history, four patients (12.1%), six patients (21.4%), and nine patients (23.1%), respectively, had focal lung lesions of high and/or intermediate significance. The false-positive rate of the CAD system was an average of 1.25 per case (range, 0 to 11).

Conclusions

Significant lung lesions are frequently missed at routine clinical interpretation of chest CT studies but may be detected if CAD is used as an additional reader.

Section snippets

Materials and Methods

One hundred patients (68 women and 32 men; mean age, 52.9 years; range, 22 to 82 years) who had undergone a CT study of the chest in one of two large academic centers were included. In all cases, the CT study had been interpreted as normal, without suspicious focal lung lesions (inclusion criterion), at routine clinical reading. Three different patient subgroups were evaluated. The first group consisted of 33 consecutive patients (26 women and 7 men; mean age, 44.4 years; range, 22 to 79 years)

Results

Within all patients, 285 image features were marked by CAD and were evaluated by the study reading panel. In 33 of the 100 patients (33%), a total of 53 lesions were found that were deemed significant by reader consensus. None of these lesions had been reported at the routine clinical reading. Five of the CAD detected lesions (9.4%) were classified as high significance, 21 lesions (39.6%) were of intermediate significance, and 27 lesions (50.9%) were of low significance. A total of 19 patients

Discussion

Two types of error can occur during the interpretation of imaging studies: perception error and classification error.2, 3, 27, 28, 29 Due to a perception error, a reader simply “does not see” an abnormality that may often be readily visible in retrospect. This can occur, for example, because of a distraction during readout, or because of a particular location of the lesion (eg, a lung nodule adjacent to a vessel of similar size). In a classification error, the observer fails to report a

References (30)

  • SuttonGC

    How accurate is computer-aided diagnosis?

    Lancet

    (1989)
  • FitzgeraldR

    Error in radiology

    Clin Radiol

    (2001)
  • BerlinL et al.

    Malpractice and radiologists in Cook County, IL: trends in 20 years of litigation

    AJR Am J Roentgenol

    (1995)
  • RenfrewDL et al.

    Error in radiology: classification and lessons in 182 cases presented at a problem case conference

    Radiology

    (1992)
  • BerlinL

    Malpractice issues in radiology: perceptual errors

    AJR Am J Roentgenol

    (1996)
  • HenschkeCI et al.

    Work-up of the solitary pulmonary nodule. American College of Radiology. ACR Appropriateness Criteria

    Radiology

    (2000)
  • GreeneRE

    Missed lung nodules: lost opportunities for cancer cure

    Radiology

    (1992)
  • GurneyJW

    Missed lung cancer at CT: imaging findings in nine patients

    Radiology

    (1996)
  • WhiteCS et al.

    Primary carcinoma of the lung overlooked at CT: analysis of findings in 14 patients

    Radiology

    (1996)
  • KakinumaR et al.

    Detection failures in spiral CT screening for lung cancer: analysis of CT findings

    Radiology

    (1999)
  • WhiteCS et al.

    Missed lung cancer on chest radiography and computed tomography: imaging and medicolegal issues

    J Thorac Imaging

    (1999)
  • LiF et al.

    Lung cancers missed at low-dose helical CT screening in a general population: comparison of clinical, histopathologic, and imaging findings

    Radiology

    (2002)
  • ShahPK et al.

    Missed non-small cell lung cancer: radiographic findings of potentially resectable lesions evident only in retrospect

    Radiology

    (2003)
  • BrownJ et al.

    Mammography screening: an incremental cost effectiveness analysis of double versus single reading of mammograms

    BMJ

    (1996)
  • GollubMJ et al.

    Clinical importance of reinterpretation of body CT scans obtained elsewhere in patients referred for care at a tertiary cancer center

    Radiology

    (1999)
  • Cited by (0)

    Drs. Peldschus, Costello, and Schoepf had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

    Dr. Peldschus is supported by the Biomedical Sciences Exchange Program, Hannover, Germany. Dr. Wood is an employee of R2 Technology Inc., Sunnyvale, CA. Dr. Costello is the recipient of a research grant provided by R2 Technology Inc., Sunnyvale, CA. Dr. Schoepf is the recipient of an unrestricted research grant provided by Siemens Corporate Research, Princeton, NJ. The sponsors had no part in the design and conduct of the study or the collection, management, analysis, and interpretation of the data.

    View full text