emma MINC到Matlab的接口（已弃用）_emma MINC to Matlab interface (depr

The following must be completed in order to implement a two-compartment rCBF model in Matlab: 1) Get the most recent articles on performing two-compartment analysis that includes solving for V0. COMPLETED ON: July 5, 1993 STATUS: Solving for V0 as well as K1 and k2 seems quite straight forward, and should be no more difficult than solving for K1 and k2. We will use the triple weighted integration method, either applied directly, or by using Hiroto's more elegant modification. Although the desired final product is a package that solves for K1, k2 and V0, it is easier to start by neglecting V0, delay, and dispersion. In order to perform the analysis while ignoring V0, the following must be completed: 2) Create a Matlab routine that generates the rR lookup table. This lookup table is used to find a value for k2 at each pixel, and should be 100x1 (100 values of rR corresponding to values of k2 from 0.01 minutes to 1 minute). COMPLETED ON: June 25, 1993 STATUS: June 29, 1993: Overhauled the function since we realized that the convolution was not being performed correctly. The function now creates evenly sampled data sets for the purpose of convolution, convolutes, and then re-samples back to the original sample spacing. We discovered that unevenly sampled data sets did not convolute correctly. June 25, 1993: Completed. June 21, 1993: Created an rR lookup table with Matlab, but not an actual .m file. However, all of the commands used were saved in a matlab diary. 3) Create a Matlab routine that calculates a value for rL. Since there is a different value of rL for each pixel in the image, rL should be 16384x1 for every slice. COMPLETED ON: June 21, 1993 STATUS: Completed. 4) Create a Matlab routine that integrates the above two routines by using the look-up table to find an approximation for k2. This routine can probably also calculate K1 since this is straight forward once k2 is known. This routine should probably just take a handle to an input data set, and a handle to an output data set. COMPLETED ON: July 2, 1993 STATUS: July 5, 1993: Worked out the correct units for K1 and k2. K1 was being expressed as: (nCi * sec * (g blood)) / ((mL tissue) * counts * sec) It should be expressed as: (mL blood) / ((g tissue) * minute) k2 was being expressed as s^-1, and should be expressed as min^-1. June 30, 1993: Routine now calculates a K1 image as well as a k2 image. However, while this image appears qualitatively similar to a K1 image produced by Jin's K1_image_auto.for, the values themselves are off by a little more than four orders of magnitude. June 25, 1993: A .m file exists, but it is very rudimentary. Data for a k2 image is generated, but no K1 image is generated. June 21, 1993: Played around with getting values for k2 by using the above two functions. No actual .m file exists yet, but once again all of the commands used were saved in a matlab diary. The above three routines will provide an rCBF package that does not take V0, delay or dispersion into account. We are currently adding delay and dispersion correction to the above routines, and from there we will move on to performing an analysis that also solves for V0. The steps for finding the delay constant delta are: 1) Find A(t) by: i) integrating a slice across its frames, and then selecting all pixels that have a value greater than 1.8 times the mean of the integrated image. ii) allow the user to interactively adjust the mask by selecting a new scale factor. iii) use pixels selected by the above procedure as a mask applied to every frame in the slice, and then calculate the mean of each image. The resulting values are used as A(t). COMPLETED ON: July 16, 1993 STATUS: July 16,1993: Finished July 9, 1993: Wrote a Matlab script that implements the above three steps and uses getmask.m. July 7, 1993: Created getmask.m. This function handles the second part of the above by displaying the image, and allowing the user to interactively adjust the scale factor. getmask returns the resulting mask as 0's and 1's. 2) Use A(t) to do a least squares fit to the blood delay and dispersion equation. COMPLETED ON: August 1, 1993 STATUS: Aug. 1, 1993: Completed. The dispersion correction problem was resolved by converting the routine currently used by the FORTRAN K1 image generation program to Matlab. The delay correction was not changed, except by using a better (trapezoidal) integration routine. Unfortunately, the curve fitting procedure is now very slow, and this must be solved. July 16,1993: After satisfying ourselves that the stepwise fitting method described below gave reasonable results, we decided to add dispersion correction. Implementation was quite easy, but unfortunately adding dispersion correction made things worse. The blood curve no longer followed the brain curve closely. July 15,1993: In order to try and solve the overdetermination problem, we have taken a stepwise incrementation approach to finding delta. We choose a value of delta, and peform the curve fit while holding delta constant. We then choose a new value of delta, and perform the three term curve fit again (for alpha, beta, and gamma). After having performed the fit for several values of delta, we choose the fit that produced the least amount of error. At the moment, we are stepping through delta with increments of 1 second. July 12,1993: Found that the curve fitting problem seems to be overdetermined. We can find several values of alpha, beta, gamma, and delta that give good curve fits. July 9, 1993: Performed the least squares fit without taking dispersion into account. The fit along the rising slope was very good, but at the top of the slope the fit function exploded. We do have some ideas for solving this problem, however. July 7, 1993: Took a first pass look at least squares fitting in general, and the matlab functions that support it in particular. In order to complete two-compartment rCBF analysis, we must: 1) Calculate the delay and dispersion constants, and apply these as corrections to the blood data used in our analysis. Please see the previous section for details on this step. 2) Perform a full two-compartment analysis that returns K1, k2, and V0 using the triple-weighted integration method. COMPLETED ON: STATUS: Aug. 1, 1993: Our implementation is currently undergoing