Deploying to demo from @ 15f92d0 🚀

niivue · Jun 12, 2024 · 1fc34f3 · 1fc34f3
1 parent e82c976
commit 1fc34f3
Show file tree

Hide file tree

Showing 12 changed files with 200 additions and 34 deletions.
diff --git a/assets/index-xAR_-jXP.js → assets/index-kXSW97rn.js b/assets/index-xAR_-jXP.js → assets/index-kXSW97rn.js
diff --git a/index.html b/index.html
@@ -6,21 +6,22 @@
   <meta name="viewport" content="width=device-width, initial-scale=1.0" />
   <link rel="stylesheet" href="./niivue.css" />
   <title>Niivue Neglect Predictions</title>
-  <script type="module" crossorigin src="./assets/index-xAR_-jXP.js"></script>
+  <script type="module" crossorigin src="./assets/index-kXSW97rn.js"></script>
 </head>
 
 <body>
   <header>
     <button id="openBtn">Open Lesion Map</button>
+    <input type="checkbox" id="cocCheck" checked />
     <label for="cocNumber">Acute CoC Score</label>
-    <input style="width: 7em" type="number" id="cocNumber" min="-1" value ="0.65" max="1" />
+    <input style="width: 7em" type="number" id="cocNumber" min="-1" value ="0.65" max="1"/>
     <button id="predictBtn">Prediction</button>
     <label for="maskSlider">Mask Opacity</label>
     <input
         type="range"
         min="1"
         max="255"
-        value="128"
+        value="64"
         class="slider"
         id="maskSlider"
     />
@@ -33,7 +34,7 @@
         class="slider"
         id="lesionSlider"
     />
-    <button id="aboutBtn">About</button>
+    <button class="btn btn-success" onclick=" window.open('https://github.com/niivue/niivue-neglect','_blank')">About</button>
   </header>
   <main id="canvas-container">
     <canvas id="gl1"></canvas>

diff --git a/mask_noCoC.nii.gz b/mask_noCoC.nii.gz
diff --git a/mat2nii.m b/mat2nii.m
@@ -2,8 +2,10 @@
 % convert model with Radial basis function kernels from matlab to NIfTI
 % we could use JSON, but we want to retain precision
 
+%fnm = 'models_5x10_diff';
+fnm = 'models_noCoC';
+models = load(strcat(fnm, '.mat')).models;
 
-models = load('models_5x10_diff.mat').models;
 %
 v = 1;
 nii(v) = numel(models);
@@ -29,4 +31,4 @@
         v = v + 1;
     end
 end
-niftiwrite(nii,'models_5x10_diff.nii')
+niftiwrite(nii,strcat(fnm, '.nii'))
diff --git a/models_noCoC.mat b/models_noCoC.mat
diff --git a/models_noCoC.nii.gz b/models_noCoC.nii.gz
diff --git a/neglect_predict.m b/neglect_predict.m
@@ -1,6 +1,6 @@
 function neglect_predict(fnm, acuteCoC)
-%Predict chronic recovery based on lesion map and acute center of
-%cancellation score.
+%Predict chronic neglect using lesion map and acute center of cancellation
+% Assumes lesion map is normalized to 181x217x181 using SPM
 %Examples
 % neglect_predict; %use GUI;
 % neglect_predict('M2095_lesion.nii.gz', 0.65)

diff --git a/neglect_predict_noCoC.asv b/neglect_predict_noCoC.asv
@@ -0,0 +1,81 @@
+function neglect_predict_noCoC(fnm)
+%Predict chronic recovery based on lesion map.
+% no acute CoC available (less accurate)
+%Examples
+% neglect_predict; %use GUI;
+% neglect_predict_noCoC('M2095_lesion.nii.gz')
+
+    if nargin < 1
+        [p_file, p_path] = uigetfile('*.nii.gz;*.nii', 'Select lesion map');
+        if p_file==0
+            return
+        end
+        fnm = fullfile(p_path, p_file);
+    end
+    lesion = niftiread(fnm);
+    lesionVolTotalML = nnz(lesion) / 1000; %convert voxels to ML
+    mpath = fileparts(mfilename("fullpath"));
+    fnmMsk = fullfile(mpath, 'mask_noCoC.nii.gz');
+    if ~exist(fnmMsk,'file')
+        error('Unable to find %s', fnmMsk)
+    end
+    maskVox = uint8(niftiread(fnmMsk) > 0);
+    nnz(lesion(maskVox > 
+    %PCA
+    fnmPCA = fullfile(mpath, 'pca_noCoC.mat');
+    if ~exist(fnmPCA,'file')
+        error('Unable to find %s', fnmPCA)
+    end
+    pca_val = load(fnmPCA).pca_val;
+    if size(pca_val.mu,2) ~= nnz(maskVox)
+        error('maskVox size does not match pca_val %d ~= %d', size(pca_val.mu,2), nnz(maskVox))
+    end
+    map_org = double(lesion(:));
+    mask_log = maskVox ~=0;
+    map = map_org(mask_log);
+    PC = (map'-pca_val.mu)*pca_val.coeff;
+    %normalize values to range 0..1
+    for i = 1:5
+        PC(i) = norm0to1(PC(i), -203.8758, 353.6057);
+    end
+    input_vector = [PC(1), PC(2), PC(3), PC(4), PC(5)];
+    fnmModel = fullfile(mpath, 'models_noCoC');
+    if ~exist(fnmPCA,'file')
+        error('Unable to find %s', fnmPCA)
+    end
+    models = load(fnmModel).models;
+    % Blank prediction-array
+    predictions_mdls = zeros(numel(models),1);
+
+    % For each single model (5-fold nested cross validation x 10 model repetitions)
+    for i = 1:numel(models)
+        % Extract information from the i-th model
+        support_vectors_i = models{i}.SVs;
+        coefficients_i = models{i}.sv_coef;
+        bias_i = -models{i}.rho;
+        % Define RBF kernel function
+        gamma_i = models{i}.Parameters(4);
+        rbf_kernel = @(x1, x2) exp(-gamma_i * sum((x1 - x2).^2));
+        % Calculate the kernel values
+        kernel_values_i = arrayfun(@(j) rbf_kernel(input_vector, support_vectors_i(j, :)), 1:size(support_vectors_i, 1));
+        % Calculate the prediction using the regression function
+        predictions_mdls(i) = sum(coefficients_i' .* kernel_values_i) + bias_i;
+        % Feature weights and bias term
+        % w = coefficients_i' * support_vectors_i;
+        % b = bias_i;
+    end
+    % Calculate mean prediction
+    prediction_mean = mean(predictions_mdls);
+    chronZ = prediction_mean * (30.1222097637568 + 1.20431863031291) - 1.20431863031291;
+    % calculate chronic CoC that can be interpreted by the user
+    chronCoC = chronZ * 0.0216 + 0.00803;
+    % output text
+    str = sprintf('Given %gml lesion, predicted chronic CoC is %g (z= %g)\n', lesionVolTotalML, chronCoC, chronZ);
+    disp(str);
+end
+function ret = norm0to1(val, mn, mx)
+    %return normalized 0..1, linearly interpolated min..max
+    range = mx - mn;
+    ret = (val - mn)/range;
+    ret = min(max(ret,0),1);
+end
diff --git a/neglect_predict_noCoC.m b/neglect_predict_noCoC.m
@@ -0,0 +1,82 @@
+function neglect_predict_noCoC(fnm)
+%Predict chronic recovery based on lesion map.
+% no acute CoC available (less accurate)
+%Examples
+% neglect_predict; %use GUI;
+% neglect_predict_noCoC('M2095_lesion.nii.gz')
+
+    if nargin < 1
+        [p_file, p_path] = uigetfile('*.nii.gz;*.nii', 'Select lesion map');
+        if p_file==0
+            return
+        end
+        fnm = fullfile(p_path, p_file);
+    end
+    lesion = niftiread(fnm);
+    lesionVolTotalML = nnz(lesion) / 1000; %convert voxels to ML
+    mpath = fileparts(mfilename("fullpath"));
+    fnmMsk = fullfile(mpath, 'mask_noCoC.nii.gz');
+    if ~exist(fnmMsk,'file')
+        error('Unable to find %s', fnmMsk)
+    end
+    maskVox = uint8(niftiread(fnmMsk) > 0);
+    lesionVolMaskML = nnz(lesion(maskVox > 0)) / 1000; %convert voxels to ML
+    %PCA
+    fnmPCA = fullfile(mpath, 'pca_noCoC.mat');
+    if ~exist(fnmPCA,'file')
+        error('Unable to find %s', fnmPCA)
+    end
+    pca_val = load(fnmPCA).pca_val;
+    if size(pca_val.mu,2) ~= nnz(maskVox)
+        error('maskVox size does not match pca_val %d ~= %d', size(pca_val.mu,2), nnz(maskVox))
+    end
+    map_org = double(lesion(:));
+    mask_log = maskVox ~=0;
+    map = map_org(mask_log);
+    PC = (map'-pca_val.mu)*pca_val.coeff;
+    %normalize values to range 0..1
+    for i = 1:5
+        PC(i) = norm0to1(PC(i), -203.8758, 353.6057);
+    end
+    PC
+    input_vector = [PC(1), PC(2), PC(3), PC(4), PC(5)];
+    fnmModel = fullfile(mpath, 'models_noCoC');
+    if ~exist(fnmPCA,'file')
+        error('Unable to find %s', fnmPCA)
+    end
+    models = load(fnmModel).models;
+    % Blank prediction-array
+    predictions_mdls = zeros(numel(models),1);
+
+    % For each single model (5-fold nested cross validation x 10 model repetitions)
+    for i = 1:numel(models)
+        % Extract information from the i-th model
+        support_vectors_i = models{i}.SVs;
+        coefficients_i = models{i}.sv_coef;
+        bias_i = -models{i}.rho;
+        % Define RBF kernel function
+        gamma_i = models{i}.Parameters(4);
+        rbf_kernel = @(x1, x2) exp(-gamma_i * sum((x1 - x2).^2));
+        % Calculate the kernel values
+        kernel_values_i = arrayfun(@(j) rbf_kernel(input_vector, support_vectors_i(j, :)), 1:size(support_vectors_i, 1));
+        % Calculate the prediction using the regression function
+        predictions_mdls(i) = sum(coefficients_i' .* kernel_values_i) + bias_i;
+        % Feature weights and bias term
+        % w = coefficients_i' * support_vectors_i;
+        % b = bias_i;
+    end
+    % Calculate mean prediction
+    prediction_mean = mean(predictions_mdls);
+    chronZ = prediction_mean * (30.1222097637568 + 1.20431863031291) - 1.20431863031291;
+    % calculate chronic CoC that can be interpreted by the user
+    chronCoC = chronZ * 0.0216 + 0.00803;
+    % output text
+    str = sprintf('Given %gml lesion (%gml in mask), predicted chronic CoC is %g (z= %g)\n', lesionVolTotalML, lesionVolMaskML, chronCoC, chronZ);
+    disp(str);
+end
+function ret = norm0to1(val, mn, mx)
+    %return normalized 0..1, linearly interpolated min..max
+    range = mx - mn;
+    ret = (val - mn)/range;
+    ret = min(max(ret,0),1);
+end
diff --git a/pca_noCoC.mat b/pca_noCoC.mat
diff --git a/pca_noCoC_coeff.nii.gz b/pca_noCoC_coeff.nii.gz
diff --git a/pca_noCoC_mu.nii.gz b/pca_noCoC_mu.nii.gz