New applications for hyperspectral imagery in the medical field are a constant area of research, as the utilization of the spectral information in hyperspectral images shows promise for early detection of disease. In “Pathological leucocyte segmentation algorithm based on hyperspectral imaging technique,” a recent paper in Optical Engineering, Yana Guan of East China Normal University and fellow authors propose a new method of automatically recognizing and classifying white blood cells.
The Shanghai researchers emphasize that the segmentation of blood samples into nucleus, cytoplasm, erythrocyte, and background is a beneficial tool in the diagnosis of blood-related disease and that fully automatic methods do not yet exist. They note that many standard segmentation methods are gray-value dependent or shape-based, and their new approach utilizes hyperspectral imagery to segment blood samples based on spectral features of the components.
In particular, they focus on the spectral information divergence (SID) algorithm, which is used to measure spectral similarity of the pixels in the sample image once reference spectra have been obtained.
They performed preliminary tests on automatic segmentation of blood samples from leukemia patients. In these tests, the authors compare the SID algorithm with the spectral angle mapper (SAM) and K-means clustering approaches and construct a confusion matrix to determine the effectiveness of each. They found that the SID accurately classifies the blood components 96% of the time.
Segmentation results for white blood cells. First row: pictures for comparison. Second row: segmentation results by SID algorithm. Third row: segmentation results by K-means algorithm. Fourth row: segmentation results by SAM algorithm. Yellow, cyan, green, and thistle represent erythrocyte, cytoplasm, nucleus, and background, respectively.
The authors recognize there is more work to be done, including the building of a spectral library for the blood system to enable automatic recognition and classification of different components, but the preliminary results are encouraging.
Co-authors are Qingli Li, Yiting Wang, Hongying Liu, and Ziqiang Zhu.
Source: Optical Engineering 51, 053202 (2012); doi: 10.1117/1.OE.51.5.053202.
Recommended by SPIE member Sarah Lane of Georgia Tech Research Institute, Electro-Optical Systems Lab.
Special Sections in Optical Engineering on Astronomy
Optical Engineering, the flagship journal of SPIE, will have two special sections on astronomy topics in August 2013.
Manuscripts are due 1 November for the special section on adaptive optics systems, guest edited by SPIE member Brent Ellerbroek of the Thirty Meter Telescope Observatory.
Manuscripts were due 1 October for a special section on ground-based and airborne telescopes and instrumentation. Guest editor is Helen Hall of NASA Ames Research Center.
Have a question or comment about this article? Write to us at email@example.com.
To receive a print copy of SPIE Professional, the SPIE member magazine, become an SPIE member.