Due to malignant focuses of some early stage cancer were too small to be shown, there is not typical malignant manifestation in primary images. The processed images with two operators expressed different typical features of pathologic changes in fo- cuses.
Active Hyperplasia. There is a light shadow near the nipple in Fig.6A. In the processed image with the glowing edges operator (Fig.6 B), the edge lines in corresponding region are irregular. In Fig.6C, these blood vessels extracted in the interest region reveal some pathological changes. Therefore, the patient was estimated likely to suffer certain malignant changes. The pathological section acquired by the operation certified that it was active hyperplasia.
Fig. 6 Images of active hyperplasia. A is the primary image, in which the shadow near the nipple is very light. B is the processed image with the glowing edges operator, in which the edge lines in the corresponding region are distorted compared with the other locations. C is the processed image with the wrapping vessels operator, in which thick, crook blood vessels are extracted in the corresponding region.
Fig. 7 Images of atypical duct hyperplasia. A is the primary image, in which there is a blood vessel passing through the light shadow region. B is the processed image with the glowing edges operator, in which the edge lines in the corresponding region are irregular compared with the other locations. C is the processed image with the wrapping vessels operator, in which blood vessels extracted in the interested region are unwonted.
Atypical Duct Hyperplasia. In the primary NIR image (Fig.7A), there is a blood vessel passing through the light shading and the borderline of the blood vessel is slightly coarse in the interest region. Fig. 10B shows the processed image with the glowing edges operator. The edge lines in corresponding region are slightly disor- dered and cannot form regular concentric circles partly. The feature in Fig.7B demon- strated that tissues in the interest region are abnormal. Fig.7C shows the processed
Extracting Pathologic Patterns from NIR Breast Images 69 image with the wrapping vessels operator. The blood vessels extracted in the inter- ested region are unwonted. According to the characters of the three images, we esti- mated that there are some pathological changes in the breast and the changes are likely to be malignant. The pathological section of the breast acquired by the opera- tion certifies that it is an indistinct duct hyperplasia, a precancerous lesion.
4 Conclusions
We have diagnosed 1466 clinical patients with NIR breast imaging. Applying the method of digital image processing to these NIR breast images, we analyzed the pri- mary and processed images by comparison. We estimated 76 cases to be cancers among 1466 clinical cases, and 73 cases were certified to be malignant tumors by the pathological examination. Without the method, only based on primary images, the possibility of false positive and false negative was high. Bases on the method, we have avoided the performance of some biopsies for benign lesions and founded some precancerous lesion or early stage cancer. It can be concluded that with the method of digital image processing, the reliability of diagnosing mammary diseases can be raised obviously.
References
2.
3.
4.
5.
6.
7.
M. Cutler.: Transillumination as An Aid in The Diagnosis of Breast Lesion. Surg Gyneco Obstet. 48 (1929) 721-730
Ohlsson B, Gundersen J, Nilsson DM.: Diaphanography: A Method for Evaluation of The Female Breast. World J Surg. 4 (1980) 701-706
Watmough DJ.: Transillumination of Breast Tissues: Factors Governing Optimal Imaging of Lesions. Radiology. 174 (1983) 89-92
Jiang H, Iftimia NV, Xu Y, Eggert JA, Fajardo LL, Klove KL.: 2002.Near-infrared Opti- cal Imaging of The Breast With Model-based Reconstruction. Acad Radiol. 9 (1995) 186- 194
J. R. Lakowicz and K. Berndt: Frequency Domain Measurement of Photon Migration in Tissue. Chem. Phys. Lett., 16 (1990) 246-252
B. Chance, et al.: Comparison of Time Resolved and Unresolved Measurements of De- oxyhaemoglobin in Brain. Proc. Natn. Acad. Sci. USA, 8 (1988) 4971-4975
Thomsen S, Tatman D.: Physiological and Pathological Factors of Human Breast Disease That Can Influence Optical Diagnosis. Ann N Y Acad Sci. 838 (1998) 171–193
1.
Comparison of Phase-Encoded and Sensitivity-Encoded Spectroscopic Imaging*
Min Huang, Songtao Lu, Jiarui Lin, Yingjian Zhan MRSI Research Group, Institute of Biomedical Engineering, Huazhong University of Science and Technology, Wuhan, 430074, China
hmrose@163.net lstsam@yofc.com
Linjiaru@hust.edu.cn zhanyingjian@hotmail.com
Abstract. MR Spectroscopic imaging plays a more and more important role in clinical application. In this paper, we make comparison on two MRSI technolo- gies and data reconstruction methods. For the conventional phase-encoded spectroscopic imaging, the data reconstruction using FFT is simple, but the data acquisition is very time consuming and thus prohibitive in clinical settings.
Sensitivity-encoded SI is a new parallel approach of reducing the acquisition time by reducing the necessary spatial encoding steps with multiple coils. It uses the distinct spatial sensitivities of the individual coil elements to recover the missing encoding information in reconstruction. Fourfold reduction in scan time can be achieved when the factor of and are both 2, with no com- promise in spectra and spatial resolution. These improvements in data acquisi- tion and image reconstruction provide a potential value of metabolic imaging using SENSE-SI as a clinical tool.
1 Introduction
MR spectroscopic imaging (MRSI, SI) is a completely noninvasive imaging method.
In contrast to magnetic resonance imaging (MRI), MRSI can present information in the form of metabolite maps, which represent not only simply anatomy but also local metabolic states or local tissue abnormalities [1]. It shows great promise for use in basic physiological research and for clinical imaging of metabolic function. SI has been proposed as a method to localize and assess brain tumors [2], multiple sclerosis, and temporal lobe epilepsy [3].
In vivo SI suffers generally from long imaging time and poor signal-to-noise ratio (SNR), because of a combination of a weak MR signal and low metabolite concentra- tions. Low SNR limits the technique’s ability to detect metabolite abnormalities in subjects. Therefore, an increase of speed and SNR is a key factor in the success of many MRSI applications.
* Grant sponsor: National Natural Science Foundation of China, Grant number: 30070226.
University Foundation of HUST, Grant number: 0101170054.
G.-Z. Yang and T. Jiang (Eds.): MIAR 2004, LNCS 3150, pp. 70-77, 2004.
© Springer-Verlag Berlin Heidelberg 2004
Comparison of Phase-Encoded and Sensitivity-Encoded Spectroscopic Imaging 71 For the conventional phase encoding spectroscopic imaging (PHASE-SI), the data reconstruction using FFT is simple. But the data acquisition is very time consuming and thus prohibitive in clinical settings [4]. In the past ten years, many approaches have been proposed and implemented for faster MRSI. Correspondingly, a lot of reconstruction methods have also been proposed. The majority of fast SI methods are based on fast MRI sequences to accelerate data sampling, in which the signal is ac- quired under a time-varying readout gradient. All signals from at least one whole slice of k-space are acquired within the TR, e.g., from a plane using echo planar spectroscopic imaging (EPSI) [5] or from a plane using spiral SI [6], and so on. Because sampled data points are not on a rectilinear grid, special reconstruction algorithms are required. EPSI reconstruction uses shift of odd and even echoes [5].
Spiral SI uses gridding FFT [7]. However, there are some restricts. EPSI requires rapidly switching gradient field and strong power provide system. For spiral method, the smoothing effects of gridding kernel will deconvolve in the image domain, thus will affect the resolution of image. And now, a new parallel imaging technique based on sensitivity-encoded (SENSE) has arisen [8-11]. SENSE-SI applies coil arrays for parallel data acquisition to reduce the acquisition time by reducing the number of sampled k-space points in the spatial frequency domain [12]. In the process of data reconstruction, it uses the distinct spatial sensitivities of the individual coil elements to recover the missing encoding information [13].
2 Theory
The collected raw signal equation in SI is given by
where is the density in the position (x,y,z) and is the decay constant of the k–th metabolite component, is the chemical shift frequency,
is the acquisition signal, i.e. the raw data.
We can see that the raw signal is possible by appropriately sampling the k-space and time domain K-space is the Fourier domain of the space (x, y), i.e. the data-sampling trajectory. For different SI technologies, there are various forms of k-space sampling schemes, which can be realized by gradient design.
Not only are the spatial distributions of the spins important but also are the spectral components with each spatial position. Spectral components can be gathered by col- lecting the time direction. The ultimate spatial distribution of spectral information of
72 M. Huang et al.
the imaged object get is m(x,y,f), which can be acquired by mathematical algorithm to the raw data