JP2019103531A - Medical image processing apparatus - Google Patents

Abstract

To provide a medical image processing apparatus capable of improving area extraction performance without extracting an inappropriate atlas even if an arrangement is similar.SOLUTION: On the basis of the positional information of a feature point analyzed by feature point analysis means 3 and luminance information as a feature amount analyzed by luminance value analysis means 4, selection means 5 selects an image case sample which corresponds to a processed image as a medical image and to which area information of a target part is previously given as an atlas, and region extraction means 6 extracts an area of the target part by using the selected image case sample.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a medical image processing apparatus that processes and displays medical images.

  In the medical image processing apparatus, in order to specify a target region in a medical image, a target region is automatically extracted by an image processing technique. However, as seen in examples of computed tomography (abbreviated as CT) and magnetic resonance imaging (abbreviated as MRI) devices, medical images can be obtained according to the physical characteristics of the imaging devices. The outlines of internal organs and the like that appear are not always clear, and region extraction may be difficult in normal image processing. Therefore, conventionally, a region extraction method using anatomical structure information in advance called an atlas has been adopted (see, for example, Patent Document 1). The atlas is an anatomical structure relating to relative positions and shapes of internal organs and the like, and is a kind of internal map data. Among them, positional information of the point which is anatomically labeled is included. Also, in general, due to structural diversity, multiple atlases are prepared for the same site.

  In a conventional medical image processing apparatus, a large number of atlases for a target site are stored in advance, and one or more atlass suitable for processed images are selected and used. For example, when each atlas is deformed and aligned with respect to the processing image, an atlas group in which the arrangement of anatomical marker point groups included in each atlas is majority similar is selected as likely. Then, based on the shape information of the target site possessed by those atlas groups, the target area in the processed image is extracted. Since region extraction can be performed with the atlas as prior knowledge, highly accurate results can be obtained.

JP, 2015-93192, A (paragraph 0043, FIG. 3)

  However, in the conventional medical image processing apparatus using an atlas, the appropriateness of the atlas with respect to the processing image is determined only by the arrangement of anatomical landmarks, and the atlas is sorted. If the atlas is properly selected, high-accuracy region extraction results can be expected, but there is a problem that the region extraction performance is degraded if an inappropriate atlas is extracted due to the similar arrangement. there were.

  The present invention has been made to solve the problems as described above, and can improve region extraction performance without extracting inappropriate atlas even when the arrangement is similar. An object of the present invention is to provide a medical image processing apparatus.

  According to the medical image processing apparatus of the present invention, a processed image input unit for inputting a processed image, a sample image input unit for inputting a plurality of sample images to which area information of a target region is attached in advance, the processed image and the sample image Feature point analysis means for analyzing feature point positions, Luminance value analysis means for analyzing brightness values of the processed image and the sample image, Position information of feature points analyzed by the feature point analysis means, and the brightness value analysis And a sorting means for sorting the sample image corresponding to the processed image based on the information of the brightness value analyzed by the means.

  According to the present invention, region extraction performance is further improved by adding image information having a luminance value to a conventional atlas element and selecting an appropriate atlas (image case sample) using information of the luminance value as well. be able to.

It is a block diagram which shows the structure of the medical image processing apparatus in Embodiment 1 of this invention. It is a flowchart figure which shows operation | movement of the medical image processing apparatus in Embodiment 1 of this invention. It is a block diagram which shows the structure of the medical image processing apparatus in Embodiment 2 of this invention. It is a flowchart figure which shows operation | movement of the medical image processing apparatus in Embodiment 2 of this invention.

Embodiment 1
FIG. 1 is a block diagram showing a configuration of a medical image processing apparatus 100 according to Embodiment 1 of the present invention. As shown in FIG. 1, the medical image processing apparatus 100 includes a processed image input unit 1, a sample image input unit 2, a feature point analysis unit 3, a luminance value analysis unit 4, a selection unit 5, and a region extraction unit 6. Ru.

  The processed image input unit 1 inputs a medical image to be processed. Although this image assumes what was image | photographed with CT apparatus, MRI apparatus, X (X-ray) X-ray imaging apparatus, an ultrasonic imaging apparatus etc., it does not limit an imaging apparatus and the means of reception in particular. As a matter of course, it is assumed that a desired target site to be extracted (here, an organ, a tumor, etc. to be treated) is shown.

  The sample image input means 2 inputs, as an atlas, a plurality of image example samples as sample images to which area information of a target region is given in advance. This image example sample is taken by an imaging device of the same type as the medical image input by the processed image input means 1, and is taken as an actual image example in which a target portion of the same type is shown. In addition, it is assumed that the region information of the target site is attached to the sample in advance by the doctors. The region of the target site is designated, for example, by labeling each pixel of the image as to whether or not it is a target. Alternatively, the contour of the area may be specified by a curve. There is no limitation on the number of input image example samples, but the larger the number is, the more likely it is that something very similar to the medical image to be processed is included, and it can be expected to lead to improvement in region extraction accuracy. On the other hand, since the calculation amount also increases, it is necessary to input an appropriate number of sheets.

  The feature point analysis unit 3 extracts feature points from the processed image input by the processed image input unit 1 and each image example sample input by the sample image input unit 2 and stores the position thereof. There is no limitation on the method of extracting feature points. For example, the corner detection method proposed by Harris et al. (C. Harris and M. Stephens, “A combined corner and edge detector”, Proc. Of 4th Alvey Vision Conference, pp. 147- The method using difference-of-gaussian as an extreme value detection method proposed by Lowe (see Low Point, “Object recognition from local scale invariant features,” Proc. Computer Vision, pp. 1150-1157 (1999), and a method called FAST (Features from Accelerated Segment Test) that accelerates corner detection using decision trees (E. Rosten and T. Drummond, “Machine Learning”. for High-speed Corner Detection ", Proc. European Conference on Computer Vision, pp. 430-443, 2006.).

  The luminance value analysis unit 4 analyzes the processed image input by the processed image input unit 1 and the luminance information of each image case sample input by the sample image input unit 2 and stores it as a feature amount. The analysis is individually performed around the entire image or each feature point extracted by the feature point analysis unit 3. If it is limited around feature points as in the latter case, computational efficiency can be expected. There is no limitation on the analysis method, but for example, a luminance histogram is generated, or once converted into luminance gradient information to generate a gradient direction histogram. The gradient is obtained by differentiating the luminance value of the image. By determining the gradient, it becomes possible to analyze as a luminance distribution pattern without being influenced by the absolute luminance value of each pixel of the image. A typical method of describing feature quantities by gradient orientation histograms is described in HOG (Histograms of Oriented Gradients) (DG Lowe, “Object recognition from local scale invariant features”, Proc. Of IEEE International Conference on Computer Vision, pp. 1150- 1157, 1999). In addition, see SURF (Speed-Upped Robust Feature) (H. Bay, T. Tuytelaars and LV Gool, "SURF: Speeded Up Robust Features", Proc. European Conference on Computer Vision, pp. 404-417, 2006. ), RIFF (Rotation-Invariant Fast Feature) (G. Takacs, V. Chandrasekhar, S. Tsai, D. Chen, R. Grzesczuk and B. Girod, “Unified Real-Time Tracking and Recognition with Rotation-Invariant Fast Features See, for example, “Proc. Conference on Computer Vision and Pattern Recognition, pp. 934-941, 2010.”.

  The sorting unit 5 receives the processed image input by the processed image input unit 1, the position information of the feature points extracted from each of the image example samples input by the sample image input unit 2, and the feature amount as the brightness analysis result. And select an image example sample having a high degree of similarity to the processed image. For example, feature points may be aligned between the processed image and each of the image example samples, and those having a small error and a high similarity of feature amounts may be selected. Further, when the feature amount can be described for each feature point, feature points may be associated between the processing image and each image example sample, and the selection may be made according to the number of the created correspondences.

  The area extraction unit 6 extracts the area of the target site from the processed image using the selected image case sample. For example, area extraction from the processed image is performed with the target area information assigned to the selected image example sample as an initial value. As an area extraction method, active contour models such as Snake (M. Kass, A. Witkin and D. Terzopoulos, "Snakes: Active Contour Models", International Journal of Computer Vision, vol. 1, no. 3, pp. 321-331, 1988.) and Level Sets (M. Sussman, P. Smereka and S. Osher, “A Level Set Approach for Computing Solutions to Incompressible Two-Phase Flow”, Journal of Computational Physics, vol. 114, pp. 146-159, 1994.), Graph Cuts (Y. Boykov and V. Kolmogorov, “An Experimental Comparison of Min-Cut / Max-Flow Algorithms for Energy Minimization in Vision”, IEEE Trans. See, for example, on Pattern Analysis and Machine Intelligence, vol. 26, no. 9, pp. 1124-1137, 2004).

  Next, the operation of the medical image processing apparatus 100 according to the first embodiment of the present invention will be described. FIG. 2 is a flowchart for explaining the operation of the medical image processing apparatus 100.

  First, steps S201 to S203 are performed before a medical image to be processed is input. In step S201, a plurality of image case samples are input by the sample image input unit 2. At this time, it is assumed that region information of the target site is added as an atlas to the image example sample as an atlas. In step S202, the feature point analysis means 3 extracts feature points from each image example sample inputted by the sample image input means 2, and then in step S203, the luminance value analysis means 4 similarly uses the sample image input means 2 The luminance value is analyzed from each input image example sample. Steps S202 and S203 may be interchanged.

  Subsequently, when the medical image to be processed is input by the processed image input unit 1 in step S204, feature points are extracted from the processed image input by the processed image input unit 1 by the feature point analysis unit 3 in step S205. In step S206, the luminance value analysis unit 4 analyzes the luminance value from the processed image input from the processed image input unit 1 in the same manner. Steps S205 and S206 may be interchanged.

  Next, in step S207, the processing unit 5 selects the processed image input by the processed image input unit 1, and the position information of the feature points extracted from each image example sample input by the sample image input unit 2, and the luminance Based on the information of the luminance value which is the feature amount as the analysis result, an image case sample having high similarity to the processed image is selected.

  Finally, in step S208, the region extraction unit 6 extracts the region of the target region from the processed image using the selected image case sample.

  As described above, the region extraction performance can be further improved by adding image information having a luminance value to the conventional atlas element and selecting an appropriate atlas (image case sample) using the information of the luminance value as well. it can.

  As described above, in the medical image processing apparatus 100 according to the first embodiment of the present invention, the position information of the feature points analyzed by the feature point analysis unit 3 and the brightness information as the feature value analyzed by the brightness value analysis unit 4 Using the image case samples selected by the area extraction unit 6 to select the image example samples to which the region information of the target region is previously given as the atlas corresponding to the processed image as the medical image by the selection unit 5 Since the region of the target portion is extracted, the region extraction performance can be further improved by selecting the appropriate atlas (image example sample) using the information of the luminance value.

  Further, by directly comparing the feature point position and the image luminance value between the processed image and each image example sample, an image example sample suitable for the processed image can be correctly sorted. Furthermore, the calculation can be made more efficient by limiting the luminance value analysis around the feature points. Also, by analyzing the luminance gradient or converting it into a histogram, it is possible to absorb the strict difference due to the individual difference of the actual image, so that the similarity can be determined robustly.

Second Embodiment
The first embodiment shows the case where the processing target as a medical image and the image case sample are individually compared, but the second embodiment shows a case where image case samples are grouped and compared.

  FIG. 3 is a block diagram showing the configuration of a medical image processing apparatus 101 according to a second embodiment of the present invention. As shown in FIG. 3, the medical image processing apparatus 101 includes a clustering unit 7 in addition to the configuration of the medical image processing apparatus 100 of the first embodiment. The other configuration of the medical image processing apparatus 101 is the same as that of the medical image processing apparatus 100 according to the first embodiment, and the corresponding parts are denoted by the same reference numerals and the description thereof will be omitted.

  The clustering means 7 receives the results of the feature point analysis means 3 and the luminance value analysis means 4, compares feature point positions and image luminance values among a plurality of image example samples, and groups similar ones of the image example samples. Do. For example, alignment of feature points is performed between image case samples, and those having a small error and high similarity of feature amounts are grouped. In addition, when the feature amount can be described for each feature point, the feature points may be associated between the respective image example samples, and grouping may be performed according to the number of the matched cases. Specific clustering methods include the k-means method (J. MacQueen, “Some Methods for Classification and Analysis of Multivariate Observations”, Proc. Of 5th Berkeley Symposium on Mathematical Statistics and Probability, Univ. Of California Press, pp. 281. -297, 1967.) or the like may be used. Further, the clustering means 7 selects one sample representative of each group after grouping.

  The sorting unit 5 selects one of the groups most similar to the processed image from the information obtained by the feature point analysis unit 3 and the luminance value analysis unit 4 between the processed image and the image example samples representing the respective groups. Sort out.

  Region extraction means 6 extracts the region of the target region from the processed image using the image case samples belonging to the selected group. For example, in each of the image example samples belonging to the selected group, area extraction from the processed image is performed with the target area information assigned to the image example samples as an initial value. After region extraction is individually performed on samples belonging to the selected group, the results are combined to obtain the final result. For the combining process, a logical product, a logical ring, an averaging process, etc. may be used. Alternatively, the target regions of the image case samples included in the selected group may be combined first to extract regions in the processed image.

  Next, the operation of the medical image processing apparatus 101 according to the second embodiment of the present invention will be described. FIG. 4 is a flowchart for explaining the operation of the medical image processing apparatus 101. Steps S401 to S403 are the same as steps S201 to S203 in the first embodiment.

  In step S404, the clustering unit 7 receives the results of the feature point analysis unit 3 and the luminance value analysis unit 4 and compares feature point positions and image luminance values among a plurality of image example samples input by the sample image input unit 2 And group similar image case samples together. Steps S405 to S407 are similar to steps S204 to S206 in the first embodiment.

  In step S408, the processing unit 5 selects between the processed image input by the processed image input unit 1 and the image example samples representing each group among the image example samples input by the sample image input unit 2; From the information obtained by the feature point analysis means and the luminance value analysis means, one group most similar to the processed image is selected.

  Finally, in step S409, the region extraction unit 6 extracts the region of the target region from the processed image using the image case samples belonging to the selected group.

  Thus, grouping is performed in advance by providing a clustering unit that compares feature point positions and image luminance values among a plurality of image example samples and groups the image example samples based on the results. The calculation in the sorting means can be made more efficient.

  As described above, in the medical image processing apparatus 101 according to the second embodiment of the present invention, position information of feature points analyzed by the feature point analysis unit 3 and luminance information as a feature amount analyzed by the luminance value analysis unit 4 Based on the above, the clustering unit 7 groups similar ones among the image case samples to which area information of the target site is given in advance as an atlas as the atlas, and the sorting unit 5 corresponds to the processed image as a medical image. The group of selected image example samples is selected, and the area of the target portion is extracted using the group of image example samples selected by the area extraction means 6. Therefore, not only the effects of Embodiment 1 can be exhibited. By grouping in advance, the calculation in the sorting means can be made more efficient.

  In the present invention, within the scope of the invention, each embodiment can be freely combined, or each embodiment can be appropriately modified or omitted.

Reference Signs List 1 processed image input means, 2 sample image input means, 3 feature point analysis means, 4 luminance value analysis means, 5 selection means, 6 area extraction means, 7 clustering means, 100, 101 medical image processing apparatus

Claims (8)

Processed image input means for inputting a processed image;
Sample image input means for inputting a plurality of sample images to which region information of a target region is given in advance;
Feature point analysis means for analyzing feature point positions of the processed image and the sample image;
Luminance value analysis means for analyzing the luminance values of the processed image and the sample image;
A sorting unit for sorting the sample image corresponding to the processed image based on position information of the feature point analyzed by the feature point analysis unit and information of a luminance value analyzed by the luminance value analysis unit; Medical image processing apparatus characterized by   The medical image processing apparatus according to claim 1, further comprising a region extraction unit that extracts a region of a target region from the processed image using the selected sample image. A clustering unit that groups the sample images based on position information of the feature points and information of the luminance value;
The group of sample images corresponding to the processed image is sorted out from the group of sample images grouped by the clustering means, and the region of the target region is selected from the processed image using the sorted group of sample images The medical image processing apparatus according to claim 2, characterized by extracting.   The medical image according to claim 3, wherein the region extraction unit extracts a region of a target region from the processed image for each of the sample images included in the group selected by the selection unit, and combines the regions. Processing unit.   The medical image processing apparatus according to claim 4, wherein the combination is an averaging process.   The medical image processing apparatus according to any one of claims 1 to 5, wherein the luminance value analysis unit analyzes the luminance value around the feature point.   The medical image processing apparatus according to any one of claims 1 to 6, wherein the luminance value analysis unit analyzes a luminance gradient.   The medical image processing apparatus according to any one of claims 1 to 7, wherein the brightness value analysis unit analyzes a brightness value or a histogram of a brightness gradient.

Images