Automated mediastinal lymph node detection from CT volumes based on intensity targeted radial structure tensor...

Area of Science:

• Medical Imaging
• Radiology
• Computer-Aided Diagnosis

Background:

• Accurate detection of mediastinal lymph nodes is crucial for diagnosing thoracic diseases.
• Conventional methods like the radial structure tensor (RST) filter struggle with lymph nodes near extreme intensity regions in CT scans.

Purpose of the Study:

• To develop an improved filter for mediastinal lymph node detection in chest CT volumes.
• To enhance the accuracy of lymph node detection by integrating intensity information.

Main Methods:

• Development of an intensity-targeted radial structure tensor (ITRST) filter.
• Implementation of a two-step detection algorithm: ITRST filtering for candidate identification and support vector machine classification for false positive removal.
• Comparative analysis with RST and Hessian filters on 47 contrast-enhanced chest CT volumes.

Main Results:

• The ITRST filter achieved a detection rate of 84.2% for lymph nodes with a short axis of at least 10 mm.
• The ITRST filter resulted in 9.1 false positives per volume.
• Performance of the ITRST filter surpassed that of the conventional RST and Hessian filters.

Conclusions:

• The proposed ITRST filter significantly improves mediastinal lymph node detection in chest CT.
• Integrating intensity prior knowledge enhances filter robustness, particularly in challenging cases.
• The ITRST filter-based algorithm offers a promising advancement for computer-aided thoracic disease diagnosis.