diff --git a/src/MATLAB/+RSF/getImf.m b/src/MATLAB/+RSF/getImf.m
index 635d6f3253ab0cc821704cb940a06711b97122cd..fbe762aa432152bcb15983905d48e1bf355dadeb 100644
--- a/src/MATLAB/+RSF/getImf.m
+++ b/src/MATLAB/+RSF/getImf.m
@@ -6,10 +6,10 @@ function imf = getImf(pData,radii)
% radii = {[0.5,8,8]}; % plate
%
-% NGPU = HIP.Cuda.DeviceCount;
+NGPU = HIP.Cuda.DeviceCount;
% p = ljsStartParallel(8*NGPU);
-N_CUDA_PROCS = 48;
+N_CUDA_PROCS = 16 * NGPU;
startTarget = 1;
imf = {};
im1 = Composite();
diff --git a/src/MATLAB/clusterExamples/goLR.m b/src/MATLAB/clusterExamples/goLR.m
index 8fa05c2dfcfaf4ddaa326d9894f2d80c7f179527..0e7d6bb75e8885f0db224c659c865d59fce873f5 100644
--- a/src/MATLAB/clusterExamples/goLR.m
+++ b/src/MATLAB/clusterExamples/goLR.m
@@ -1,7 +1,8 @@
tic
-ROOT = '/g/leverjs/Schuman_OCT/OCT/qualified';
+ROOT = '/g/leverjs/Schuman_OCT/OCT/qualifiedYes';
PID = 'P10010';
+% PID = 'P10125'
flist = dir(fullfile(ROOT,['/' PID '*.LEVER']));
mx = regexp({flist.name},'P(?<PID>\d+)_(?<scanType>.+)_(?<date>\d+-\d+-\d+)_(?<time>\d+-\d+-\d+)_(?<eye>\w\w)_(?<scanID>.+?)_.*.LEVER','names');
tblMeta = struct2table(vertcat(mx{:}));
@@ -15,10 +16,10 @@ tblMeta = tblMeta(contains(tblMeta.scanType,'Optic'),:);
p = ljsStartParallel(96);
-% radii = {[0.5,8,8],[5,5,5],[0,0,0]}; % plate,blob
+radii = {[0.5,8,8],[5,5,5],[0,0,0]}; % plate,blob
% radii = {[0.5,8,8]}; % plate
-radii = {[3]} % dark tubes
+% radii = {[3]} % dark tubes
imf = RSF.getImf(tblMeta,radii);
d = [];
parfor i = 1:length(imf)
diff --git a/src/MATLAB/importMetadata/readme.txt b/src/MATLAB/importMetadata/readme.txt
new file mode 100644
index 0000000000000000000000000000000000000000..ea28fd6019e6afa2fb12d3ca649436b25e9fb30f
--- /dev/null
+++ b/src/MATLAB/importMetadata/readme.txt
@@ -0,0 +1,15 @@
+Hi Andy,
+
+Sorry for the messy spreads. But yes, those would be the files which contain the MD used in classification (column R).
+And yes the Project IDs should be associated with the image files.
+This data probably needed some tidying up before being shared but if you would still like to work on it its important to make sure that we use tests with
+• threshold of Visual Field 24-2 Test Pattern (column I)
+• have a qualification_status of yes (column CE)
+• have < 33% average false_positive_p, false_negative_p, fixation_loss_c_p (columns F,G,H)
+
+In regards to multiple MDs on the same visit, besides the rows that are outright duplicates, please select the tests which have the lower total reliability errors (columns F,G,H).
+
+I hope that clarifies some of the issues, in the meantime I will work on resending a cleaner dataset!
+
+Best,
+Ronald
diff --git a/src/MATLAB/ncdVSvfmd/goKernelCorr.m b/src/MATLAB/ncdVSvfmd/goKernelCorr.m
index 9451fd99008afe9d21a57c6677a9d6984305829c..caa95c1956ceb1bc4c630efc3c0e032398d3c3fd 100644
--- a/src/MATLAB/ncdVSvfmd/goKernelCorr.m
+++ b/src/MATLAB/ncdVSvfmd/goKernelCorr.m
@@ -1,7 +1,11 @@
+SAVE_FOLDER = './results';
+if ~exist("SAVE_FOLDER",'dir')
+ mkdir('SAVE_FOLDER');
+end
datetime
kernelCorrStart = tic();
eye = 'OD';
-ROOT = '/g/leverjs/Schuman_OCT/OCT/qualified';
+ROOT = '/g/leverjs/Schuman_OCT/OCT/qualifiedYes';
% scanType = 'Macular Cube 200x200';
scanType = 'Optic Disc Cube 200x200';
@@ -12,11 +16,15 @@ tblMeta = getMetaOCT(ROOT,eye,scanType);
load('../qualify/onh_rnfl_thickness.mat');
-radii = {[0.5,8,8],[5,5,5],[4]}; % plate,blob
-% radii = {[0.5,8,8]}; % plate
+% radii = {[0.5,8,8],[5,5,5],[8]}; % plate,blob
+% radii = {[0.5,8,8],[5,5,5],[0]}; % plate,blob
+radii = {[0.5,8,8]}; % plate
[rho2,tblVelo, pData, dxx] = kernelCorr(radii,tblVFMD,tblMeta,tblRNFL);
[rmse1,~] = regnetCorr(tblVelo);median(rmse1)
+
+saveFile = fullfile(SAVE_FOLDER,['qualYes_ncdVSvfmd_' jsonencode(radii) '.mat']);
+save(saveFile)
diff --git a/src/MATLAB/ncdVSvfmd/regnetCorr.m b/src/MATLAB/ncdVSvfmd/regnetCorr.m
index 6a0193101b72460b76db8663e57329935b69d3c4..553e7a06ebcc830b29c00df25dc28bbcdc728b61 100644
--- a/src/MATLAB/ncdVSvfmd/regnetCorr.m
+++ b/src/MATLAB/ncdVSvfmd/regnetCorr.m
@@ -7,7 +7,7 @@ function [err,mdlX] = regnetCorr(tblVelo)
% X = [tblVelo.ncd,tblVelo.md];
t2 = tblVelo(tblVelo.dmd~=0,:);
-X1 = [t2.ncd];
+X1 = [t2.ncd,t2.deltaRNFL];
X2 = [t2.ncd,t2.md];
T = t2.dmd;
% T = t2.deltaRNFL;
@@ -27,8 +27,11 @@ for i = 1:length(xTarget)
mdlX{i} = mdl;
pred = mdl.predict(xTarget(i,:));
err(i) = abs(T(i) - pred);
- [i,median(err)]
+ fprintf(1,'%d,%0.2f, ',i,median(err));
+ drawnow;
+ if mod(i,10) == 0,fprintf(1,'\n');end
end
+fprintf(1,'\n');
% mean(err)
4;
diff --git a/src/MATLAB/qualify/goQualify.m b/src/MATLAB/qualify/goQualify.m
index 1c94df3efb2b888aeda85e534a7f088733282e88..594785942f51beafc456ecb6c721ef4c66c559ba 100644
--- a/src/MATLAB/qualify/goQualify.m
+++ b/src/MATLAB/qualify/goQualify.m
@@ -1,13 +1,13 @@
ROOT = '/g/leverjs/Schuman_OCT/OCT/combined';
-target = '/g/leverjs/Schuman_OCT/OCT/qualified';
+target = '/g/leverjs/Schuman_OCT/OCT/qualifiedYes';
if ~exist(target,'dir')
mkdir(target);
end
-% flist = dir(fullfile(ROOT,'**/*.LEVER'));
-% qONH = readtable('nyu onh matched.xlsx');
-% qMacula = readtable('nyu mac matched.xlsx');
-% [~,fONH,~] = fileparts(qONH.file_name);
-% [~,fMacula,~] = fileparts(qMacula.file_name);
+flist = dir(fullfile(ROOT,'**/*.LEVER'));
+qONH = readtable('nyu onh matched.xlsx');
+qMacula = readtable('nyu mac matched.xlsx');
+[~,fONH,~] = fileparts(qONH.file_name);
+[~,fMacula,~] = fileparts(qMacula.file_name);
qlist = [];
for i = 1 : length(flist)
@@ -23,11 +23,10 @@ for i = 1 : length(flist)
continue
end
qs = lower(tidx.qualification_status);
- if any(strcmp(qs,'no'))
- continue
+ if any(strcmp(qs,'yes'))
+ qlist = [qlist,flist(i)];
+ src = fullfile(flist(i).folder,fname);
+ cmd = ['cp "' src '".* ' target];
+ system(cmd);
end
- qlist = [qlist,flist(i)];
- src = fullfile(flist(i).folder,fname);
- cmd = ['cp "' src '".* ' target];
- system(cmd);
end
diff --git a/src/MATLAB/qualify/onh_rnfl_thickness.mat b/src/MATLAB/qualify/onh_rnfl_thickness.mat
new file mode 100644
index 0000000000000000000000000000000000000000..e451b31e5ca3a558ead7fc5d1ee0ef2e014b4367
Binary files /dev/null and b/src/MATLAB/qualify/onh_rnfl_thickness.mat differ