为什么 Matlab 的 for 循环在迭代次数多的情况下速度慢，而在迭代次数少的情况下速度快？

Question

我正在运行ning 一个函数来从 100,000 多个患者 xray dicom 文件中提取一些信息。出于安全目的，文件存储在 veracrypt 加密容器中。

当我运行函数在一小部分文件上执行时它执行得非常快，但是当我运行函数在整个数据集上时相比之下它非常慢，从几个文件每秒到每秒 1 个文件（大约）。

我在想为什么会这样？我曾尝试将数据存储在 ssd 和普通硬盘驱动器上，当使用较大的数据集与较小的数据集相比时，速度会变慢。

我已经添加了下面的代码以供参考，但还没有完全评论它..这是我的论文，所以我会在提取完成后这样做..

感谢您的帮助。

function [ DB, corrupted_files ] = extract_from_dcm( folder_name )
%EXTRACT_FROM_DCM Summary of this function goes here
%   Detailed explanation goes here

if nargin == 0
    folder_name = 'I:\Find and Treat\MXU Old Backup05';
end

Database_Check = strcat(folder_name, '\DataBase.mat');

if exist(Database_Check, 'file')

    load(Database_Check);
    entry_start = length(DB) + 1;

else

    entry_start = 1;
    [ found_dicoms ] = recursive_search( folder_name );

end

mat_file_location = strcat(folder_name, '\DataBase.mat');
excel_DB_file = strcat(folder_name, '\DataBase.xlsx');
excel_Corrupted_file = strcat(folder_name, '\Corrupted_Files.xlsx');

% the recursive search creates a struct with the path for each
% dcm file found. the list is then recursivly used to locate
% the image and extract the relevant information from it.


fprintf('---------------------------------------------\n');
fprintf('Start Patient Data Extraction\n');
tic
h = waitbar(0,'','Name','Patient Data Extraction');
entry_end = length(found_dicoms);

if entry_end == 0
    %     set(handles.info_box, 'String', 'No Dicom Files Found in this Folder or its Subfolders');
else
    %     set(handles.info_box, 'String', 'Congratulations Dicom Files have been found Look Through the Data Base using the Buttons Below....Press Save Button to save the Database. (Database Save format is EXCEL SpreadSheet and MAT file');

    for  kk = entry_start : entry_end
        progress = kk/entry_end;
        progress_percent = round(progress * 100);
        waitbar(progress,h, sprintf('%d%%   %d/%d of images processed', progress_percent, kk, entry_end));

        img_full_path = found_dicoms(kk).name;

        %         search_path = folder_name;
        %         img_full_path = strrep(img_full_path, search_path, '');

        try              %# Attempt to perform some computation
            dicom_info = dicominfo(img_full_path); %# The operation you are trying to perform goes here

            try              %# Attempt to perform some computation
                dicom_read = dicomread(dicom_info); %# The operation you are trying to perform goes here

                old = dicominfo(img_full_path);

                DB(kk).StudyDate = old.StudyDate;
                DB(kk).StudyTime = old.StudyTime;


                if isfield(old.PatientName, 'FamilyName')
                    DB(kk).Forename = old.PatientName.FamilyName;
                else
                    DB(kk).Forename = 'NA';
                end

                if isfield(old.PatientName, 'GivenName')
                    DB(kk).LastName = old.PatientName.GivenName;
                else
                    DB(kk).LastName = 'NA';
                end

                if isfield(old, 'PatientSex')
                    DB(kk).PatientSex = old.PatientSex;
                else
                    DB(kk).PatientSex = 'NA';
                end

                if isempty(old.PatientBirthDate)
                    DB(kk).PatientBirthDate = '00000000';
                else
                    DB(kk).PatientBirthDate = old.PatientBirthDate;
                end

                if strcmp(old.Manufacturer, 'Philips Medical Systems')
                    DB(kk).Van = '1';
                else
                    DB(kk).Van = '0';% section to represent organising by different vans
                end

                DB(kk).img_Path = img_full_path;

                save(mat_file_location,'DB','found_dicoms');

            catch exception  %# Catch the exception
                fprintf('read - file %d corrupt.\n',kk);
                continue       %# Pass control to the next loop iteration
            end



        catch exception  %# Catch the exception
            fprintf('info - file %d corrupt.\n',kk);
            continue       %# Pass control to the next loop iteration
        end


    end

end


[ corrupted_files, DB ] = corruption_check( DB, found_dicoms, folder_name );


toc
fprintf('End Patient Data Extraction\n');
fprintf('---------------------------------------------\n');

fprintf('---------------------------------------------\n');
fprintf('Start Saving Extracted Data \n');
tic


save(mat_file_location,'DB','corrupted_files','found_dicoms');

if isempty(DB)
    msg = sprintf('No Dicom Files Found');
    msgbox(strcat(msg));
else
    DB_table = struct2table(DB);
    writetable(DB_table, excel_DB_file);
end

close(h);

toc
fprintf('End Saving Extracted Data \n');
fprintf('---------------------------------------------\n');

end

Answer 1

好的谢谢大家的帮助..

我的问题是每次迭代结束时的保存，但最大的问题是我运行 dicomread 函数所在的行。我将保存更改为每处理 20 张图像发生一次。

我还删除了评论中建议的预分配，以查看在没有 dicromread 和保存膨胀的情况下它有什么不同。它比预分配慢得多。

...我只需要找到 dicomread 的解决方案（我用它来检查文件是否损坏）。

为什么 Matlab 的 for 循环在迭代次数多的情况下速度慢，而在迭代次数少的情况下速度快？

why is Matlab slow in for loop with large number of iterations but fast with a small number of iterations?

performance

matlab

for-loop

data-extraction

slowdown