近年腦腫瘤發病率呈上升趨勢，約占全身腫瘤的5%，占兒童腫瘤的70%。CT、MRI等多種影像檢查方法可用于檢測腦腫瘤，其中MRI應用于腦腫瘤成像效果最佳。精準的腦腫瘤分割是病情診斷、手術規劃及后期治療的必備條件，既往研究者對腦部腫瘤分割算法進行了深入研究，并取得了很多成果。然而腦部結構復雜，包括腦皮層、灰質、白質、胼胝體、腦脊液等組織，分割精度難以保證。目前臨床使用最廣泛的腦部腫瘤分割方法是模糊Ｃ均值算法和均值漂移算法。圖像分割主要包括濾波和分割兩部分，一般選取常用于腦部膠質瘤圖像分割的非局部均值濾波、中值濾波、各向異性濾波3種濾波方法和分水嶺算法、模糊C均值算法等常用的不同類型分割算法。

鑒于此，本項目采用傳統的圖像處理算法腦部磁共振成像腫瘤圖像進行分割，運行環境為MATLAB 2018。

function diff_im = anisodiff(im, num_iter, delta_t, kappa, option)
fprintf('Removing noise\n');fprintf('Filtering Completed !!');% Convert input image to double.
im = double(im);% PDE (partial differential equation) initial condition.
diff_im = im;% Center pixel distances.
dx = 1;
dy = 1;
dd = sqrt(2);% 2D convolution masks - finite differences.
hN = [0 1 0; 0 -1 0; 0 0 0];
hS = [0 0 0; 0 -1 0; 0 1 0];
hE = [0 0 0; 0 -1 1; 0 0 0];
hW = [0 0 0; 1 -1 0; 0 0 0];
hNE = [0 0 1; 0 -1 0; 0 0 0];
hSE = [0 0 0; 0 -1 0; 0 0 1];
hSW = [0 0 0; 0 -1 0; 1 0 0];
hNW = [1 0 0; 0 -1 0; 0 0 0];% Anisotropic diffusion.
for t = 1:num_iter% Finite differences. [imfilter(.,.,'conv') can be replaced by conv2(.,.,'same')]nablaN = imfilter(diff_im,hN,'conv');nablaS = imfilter(diff_im,hS,'conv');   nablaW = imfilter(diff_im,hW,'conv');nablaE = imfilter(diff_im,hE,'conv');   nablaNE = imfilter(diff_im,hNE,'conv');nablaSE = imfilter(diff_im,hSE,'conv');   nablaSW = imfilter(diff_im,hSW,'conv');nablaNW = imfilter(diff_im,hNW,'conv'); % Diffusion function.if option == 1cN = exp(-(nablaN/kappa).^2);cS = exp(-(nablaS/kappa).^2);cW = exp(-(nablaW/kappa).^2);cE = exp(-(nablaE/kappa).^2);cNE = exp(-(nablaNE/kappa).^2);cSE = exp(-(nablaSE/kappa).^2);cSW = exp(-(nablaSW/kappa).^2);cNW = exp(-(nablaNW/kappa).^2);elseif option == 2cN = 1./(1 + (nablaN/kappa).^2);cS = 1./(1 + (nablaS/kappa).^2);cW = 1./(1 + (nablaW/kappa).^2);cE = 1./(1 + (nablaE/kappa).^2);cNE = 1./(1 + (nablaNE/kappa).^2);cSE = 1./(1 + (nablaSE/kappa).^2);cSW = 1./(1 + (nablaSW/kappa).^2);cNW = 1./(1 + (nablaNW/kappa).^2);end% Discrete PDE solution.diff_im = diff_im + ...delta_t*(...(1/(dy^2))*cN.*nablaN + (1/(dy^2))*cS.*nablaS + ...(1/(dx^2))*cW.*nablaW + (1/(dx^2))*cE.*nablaE + ...(1/(dd^2))*cNE.*nablaNE + (1/(dd^2))*cSE.*nablaSE + ...(1/(dd^2))*cSW.*nablaSW + (1/(dd^2))*cNW.*nablaNW );完整代碼：https://mbd.pub/o/bread/mbd-ZJacmJ9s        
end

工學博士，擔任《Mechanical System and Signal Processing》《中國電機工程學報》《控制與決策》等期刊審稿專家，擅長領域：現代信號處理，機器學習，深度學習，數字孿生，時間序列分析，設備缺陷檢測、設備異常檢測、設備智能故障診斷與健康管理PHM等。