Digital synthesis of lung nodules

Area of Science:

• Radiology
• Medical Imaging
• Computer-Aided Diagnosis

Background:

• Evaluating observer accuracy for pulmonary nodule detection is crucial.
• Current methods often use nodules from clinical practice, which may lack optimal size or location for testing.
• A standardized method for creating realistic pulmonary nodules is needed to enhance study design.

Purpose of the Study:

• To describe a method for generating computer-synthesized pulmonary nodules on digitized chest radiographs.
• To assess the realism of these synthesized nodules by comparing them to actual pulmonary nodules.

Main Methods:

• A computer-generated nodule was placed onto digitized chest radiographs.
• These synthesized nodule images were paired with digitized images of real pulmonary nodules.
• Thirteen radiologists performed a two-alternative forced-choice test to differentiate synthesized from real nodules.

Main Results:

• Radiologists could only distinguish the synthesized nodules from real nodules in 51% of cases.
• This high degree of indistinguishability suggests strong visual similarity between synthesized and real nodules.

Conclusions:

• Computer-generated nodules can effectively simulate real pulmonary nodules.
• This technique offers a valuable tool for future studies on pulmonary nodule detection and observer performance.