MRI PROTOCOL FOR SEGMENTATION OF AN IMAGE DETAIL USING IMAGES ACQUIRED AT TWO DIFFERENT MAGNETIC FIELD STRENGTHS

Abstract

A magnetic resonance imaging protocol includes an acquisition segment to control an acquisition sequence to acquire magnetic resonance signals at a lower main magnetic field strength. A reconstruction segment controls reconstruction of a diagnostic magnetic resonance image from the magnetic resonance signals at a lower main magnetic field strength. A segmentation segment to control segmentation of a pre-determined image-detail of the diagnostic magnetic resonance image. In the magnetic resonance imaging protocol: the acquisition sequence has a set of imaging parameters that cause the image quality of the diagnostic magnetic resonance to be similar to the image quality of the magnetic resonance training images. The segmentation segment comprises: an initialisation portion which controls (i) access to a set of magnetic resonance training images acquired at main magnetic field of a higher main magnetic field strength (ii) registration of the diagnostic magnetic resonance image to one or more of the magnetic resonance training images and (iii) a segmentation proper applied to the diagnostic image to segment the pre-determined detail from the registered diagnostic magnetic resonance image. The one or more magnetic resonance training images includes an image detail corresponding to the pre-determined image detail in the diagnostic magnetic resonance image. Notably, the magnetic resonance training magnetic resonance images are acquired at a high magnetic field strength at 7 T and its level of detail facilitates the accurate segmentation of notably the hippocampus from diagnostic magnetic resonance images at lower field strength of 3 T. The diagnostic magnetic resonance images are acquired such that they resemble the training magnetic resonance images.

Claims

1. A magnetic resonance imaging protocol including comprising: an acquisition segment to control an acquisition sequence to acquire magnetic resonance signals at a lower main magnetic field strength, a reconstruction segment to control reconstruction of a diagnostic magnetic resonance image from the magnetic resonance signals at a lower main magnetic field strength and a segmentation segment to control segmentation of a pre-determined image-detail of the diagnostic magnetic resonance image, in which magnetic resonance imaging protocol: the acquisition sequence has a set of imaging parameters that cause the image quality of the diagnostic magnetic resonance to be similar to the image quality of magnetic resonance training images acquired at main magnetic field of a higher main magnetic field strength and the segmentation segment comprises an initialisation portion which controls access to the set of magnetic resonance training images and, registration of the diagnostic magnetic resonance image to one or more of the magnetic resonance training images, the one or more magnetic resonance training images including an image detail corresponding to the pre-determined image detail in the diagnostic magnetic resonance image and the segmentation segment comprises a segmentation proper applied to the diagnostic image to segment the pre-determined detail from the registered diagnostic magnetic resonance images.

2. The magnetic resonance imaging protocol of claim 1, wherein the acquisition sequence is a 3D multi-echo single shot turbo-spin echo sequence, wherein: the repetition time TR is in the range of 2000-3000 ms, the echo time is in the range of 200-250ms, the echo train length has 150-170 echoes and echo spacing per shot in the range of 4.4/720ms to 5.2/880ms, and which acquisition sequence performs a half-Fourier scanning of k-space in the phase-encode direction with a half-Fourier factor in the range of 0.60 to 0.73,

3. The magnetic resonance imaging protocol of claim 2, wherein a parallel imaging acquisition method is employed at a sampling reduction factor in the range of 1.3-1.5 in a first phase-encoding direction and a sampling reduction factor in the range of 1.8-2.2 in a second phase-encoding direction.

4. The magnetic resonance imaging protocol of claim 2, wherein the repetition time TR is substantially 2200 ms or 2400 ms.

5. The magnetic resonance imaging protocol of claim 2, wherein the echo time is substantially 211 ms or 226 ms.

6. The magnetic resonance imaging protocol of claim 2, wherein the echo spacing per shot is substantially 4.8/804 ms.

7. The magnetic resonance imaging protocol of claim 2, wherein the half-Fourier factor is substantially 0.675.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0039] FIG. 1 shows a schematic representation of a magnetic resonance imaging protocol of the invention.

[0040] FIG. 2 shows an example of the segmentation of hippocampus sub-structures, left side of the brain hemisphere. Three views that show the segmentation labels superimposed on the VOI of the 3 T brain MRI.

[0041] FIG. 3 shows an example of the segmentation of hippocampus sub-structures, right side of the brain hemisphere. Three views that show the segmentation labels in different colours superimposed on the VOI of the 3 T brain MRI.

DETAILED DESCRIPTION OF THE EMBODIMENTS

[0042] FIG. 1 shows a schematic representation of a magnetic resonance imaging protocol of the invention. The magnetic resonance signals for the diagnostic magnetic resonance imaging method are acquired in the acquisition segment 10 which is formed by a

[0043] MR acquisition sequence that is translated from the MR acquisition sequence that is employed to generate the training magnetic resonance images from magnetic resonance signal acquired at a higher main magnetic field strength as compared to the field strength at which the magnetic resonance signals for the diagnostic image are acquired. The translation of the MR acquisition sequence at ultra-high (e.g. 7 T) to high (e.g. 3 T) or medium (1.5 T) field strengths achieves that the acquired magnetic resonance signal give rise to a similar image contrast in both the diagnostic and training magnetic resonance images. By way of a reconstruction segment 20, the diagnostic magnetic resonance image is reconstructed from the acquired magnetic resonance signal from the acquisition segment 10. This reconstruction is known per se e.g. in the form of in (inverse) fast Fourier transform. This reconstructed diagnostic magnetic resonance image is input into the segmentation segment 30. The segmentation segment 30 includes various stages. First an initialisation stage 31 is applied in which first a global registration of a training magnetic resonance image is made to the diagnostic image, after which the volume-of-interest, e.g. a volume including the hippocampus, is extracted and again a global registration of the corresponding volumes-of-interest in the training and diagnostic images is made. Good results are achieved as follows: 1. A global and local full brain registration of the training and diagnostic images—actually whole brain volumes, 2. Extract the VOIs in the training and diagnostic images, 3. Repeat the global and local registration between the VOIs. Based on these registered volumes-of-interest a segmentation proper stage 32 is applied to the more refined details in the volume-of-interest, e.g. the anatomical details of the hippocampus. The segmentation proper includes an active deformable surface method that combines a data term represented by a Chan-Vese term plus a prior term in terms of image moments. The role of this method is to improve on the initial segmentation provided by the initialization process; the latter deforms the annotated sub-structures, via the registration method, to the regions close to that of their correct positions in the diagnostic magnetic resonance image of the brain. For example, the full brain registration is realized by the Elastix image registration software between the training magnetic resonance image that includes the annotation set obtained from atlases (T.sub.2-weighted 7 T MRI brain volume) and a r target brain volume (T.sub.2-weighted 3 T MRI brain volume) of the diagnostic magnetic resonance image. This registration is done in two steps: (i) global affine transformation, and (ii) local registration based on B-splines. The VOI extraction is such that it uses a binary mask (all sub-structures with the intensity of 255) and its envelope plus a region of “security” made up by M voxels in each cardinal direction. The VOIs for the 7 T and 3 T volumes are registered between themselves by applying again the Elastix method.

[0044] Thus, the accurate annotations in that each hippocampus sub-structure has a label and constant intensity (grey level) values and segmented anatomical details in the atlas of training magnetic resonance images is accurately transferred into a segmentation and annotation in the output diagnostic image 40 from the segmentation segment 30.

[0045] As to the atlas 50 of training magnetic resonance images, these are generated from a set of e.g. a dozen or more, ultra-high field magnetic resonance images. For these training magnetic resonance images, at ultra-high, e.g. 7 T or more, main magnetic field strength magnetic resonance signals are acquired 40 from several volunteers. From these magnetic resonance signals the training magnetic resonance images are reconstructed 61 and annotated 62 by a group of radiological experts. This leads to an atlas 70 of annotated high-diagnostic quality (contrast resolution) training magnetic resonance images. These training images include an accurate and reliable segmentation and annotation of fine anatomical details such as details of the anatomical structure of the hippocampus. For example, the training data is made up of ten annotated 7 T MRI brains. Each brain is from another volunteer. Using the volumes of each substructure a set of parameters corresponding to their geometrical moments; these parameters are the training data.

[0046] FIG. 2 shows an example of the segmentation of hippocampus sub-structures, left side of the brain hemisphere. Three views that show the segmentation labels superimposed on the VOI of the 3 T brain MRI. The hippocampus head is annotated ‘hd, the tail as ‘t1 and the dentate gyrus ‘dg’.

[0047] FIG. 3 shows an example of the segmentation of hippocampus sub-structures, right side of the brain hemisphere. Three views that show the segmentation labels in different colours superimposed on the VOI of the 3 T brain MRI. The hippocampus head is annotated ‘hd, the tail as ‘t1 and the dentate gyrus ‘dg’ and the entorhinal cortex as ‘ec’.