SKin cancer detection using ANN

I am trying to do a classification of skin cancer using ANN. I did the training of network. But i m stuck with the classification part. my objective is, first train the network using known values. after that skin data is given to network so that ANN classifies the data into cancerous or non-cancerous indicated by 1 and 0. My doubt is, if this above mentioned problem can be solved using matlab. if u guys have any idea about this, pls help me

Answer 1

[Image Analyst]

I'd say it has a reasonably high probability of accurately predicting the cancerous state, given a clever enough algorithm. Are you going to invent one on your own, or use one that has already been developed (see Vision Bib)?

Answer 2

[Walter Roberson]

Are your images those of pathology slides? Or are the images taken in infrared? If the answer to both of those is "no", then NO, the problem cannot be solved using MATLAB. There is too much overlap in visible-light appearance between cancerous moles and non-cancerous moles. (Though you can get a hint by looking for hairs on the mole.)

[ASWIN R B]

sir, i have taken dermoscopic images for classification. it is in digital format- jpeg. I have extracted features from this images using glcm in matlab. I m doing classification using the obtained features using neural network

[Walter Roberson]

Dermoscopic images are visible light, so NO, you are not going to be able to solve the problem with MATLAB (or any other programming language.)

If you choose to proceed anyhow, you can probably only expect in the order of 72% classification rate.

[rajitha kantheti]

can I use .svs or .tiff files of pathology slides...if so, how could I proceed,to detect if the slide is cancerous or not?

[Muhammad Zeeshan Ahmed Khan]

what if we apply some morphology operation to classify or segment and then hardening the image

[Walter Roberson]

Are morphology operations able to somehow locate information from outside the visible spectrum? If not, then you are not going to be able to solve the problem: cancerous moles look too much like non-cancerous moles in visible light for you to be able to distinguish between them.

Answer 3

[Greg Heath]

How many features?

How many class 1?

How many class 0?

help patternnet

doc patternnet

Search the NEWSGROUP and ANSWERS for patternnet.

To find patternnet examples

help nndatasets

Write back with the specific quesions if you have trouble with the classification/pattern-recognition examples.

Hope this helps.

Greg

[ASWIN R B]

there are six features. 2 classes- cancerous and non cancerous indicated by 1 and 0. if i could get a neural network training using GA, it will be very helpful. I tried to train with the program which i gt from matlab answers. but it is nt working properly.. Sir, could u pls give me the program for this

[Walter Roberson]

Sounds like a school project for me. Greg is not likely to give you the code for a school project.

However, if you show what you have already tried and show the error messages, and the size() and class() of all of the variables, then someone might be able to explain what the error messages mean.

[Greg Heath]

Did you search both NEWSGROUP and ANSWERS for patternnet?

Which MATLAB nndataset did you practice on?

Where is your reply with code, error messages and questions?

[ASWIN R B]

the program i used is given below:

clc

function mse_calc = mse_test(x, net, inputs, targets)

net = setwb(net, x);

y = net(inputs);

mse_calc = sum((y-targets).^2)/length(y);
end % inputs f or the neural net

load input.txt
inputs = input;

load target1.txt
targets = target1;

n = 10;

net = feedforwardnet(n);

net = configure(net, inputs, targets);

h = @(x)mse_test(x, net, inputs, targets);

ga_opts = gaoptimset('TolFun', 1e-8,'display','iter');

[x_ga_opt, err_ga] = ga(h, 3*n+1, ga_opts);

the following errors are shown when this program is run

??? Input argument "net" is undefined.

Error in ==> mse_test1 at 3
net = setwb(net, x);

Error in ==> validate>@(x)fitness(x,FitnessFcnArgs{:}) at 136
    fitness = @(x) fitness(x,FitnessFcnArgs{:});

Error in ==> fcnvectorizer at 14
            y(i,:) = feval(fun,(pop(i,:)));

Error in ==> makeState at 47
        Score =
        fcnvectorizer(state.Population(initScoreProvided+1:end,:),FitnessFcn,1,options.SerialUserFcn);

Error in ==> gaunc at 41
state =
makeState(GenomeLength,FitnessFcn,Iterate,output.problemtype,options);

Error in ==> ga at 291
        [x,fval,exitFlag,output,population,scores] =
        gaunc(FitnessFcn,nvars, ...

Error in ==> genetic at 36
[x_ga_opt, err_ga] = ga(h, 3*n+1, ga_opts);

Caused by:
    Failure in user-supplied fitness function evaluation. GA cannot
    continue.

[Walter Roberson]

You show the code for mse_test but not for mse_test1 which is the routine that has the problem.

[Greg Heath]

Walter,

It would help, immensely, if neural net questions were based on data from help nndataset examples and not the questioners own data.

Greg

[Greg Heath]

Aswin

http://www.mathworks.com/matlabcentral/answers/57609-which-network-can-i-use-for-cancer-diagnosis

[Greg Heath]

Aswin:

Try the cancer_dataset example

help cancer_dataset

doc cancer_dataset

[hamed abdulaziz]

Hi Aswin,

Did you done Skin cancer detection using ANN,please could you send the MATLAB code that you used it because of my research on the same subject,thanks in advance

Answer 4

[Mustapha Ibrahim R.]

Am undergoing a research on comparison of Medical image segmentation what should be an appropriate objective to that project

[Image Analyst]

Cite and explain the different methods. List their claimed accuracies (false positive, true positive, false negatives, false positive). Not sure what your research entails. If you want to characterize actual skin lesions in collaboration with your dermatologist or oncologist, then you might pick one or two of the "best" algorithms and code them up and try them on your own subjects/images.