This blog posts explains how to train a deep learning lymphoma sub-type classifier in accordance with our paper “Deep learning for digital pathology image analysis: A comprehensive tutorial with selected use cases”.
Please note that there has been an update to the overall tutorial pipeline, which is discussed in full here.
This text assumes that Caffe is already installed and running. For guidance on that you can reference this blog post which describes how to install it in an HPC environment (and can easily be adopted for local linux distributions).
The NIA curated this dataset to address the need of identifying three sub-types of lymphoma: Chronic Lymphocytic Leukemia (CLL), Follicular Lymphoma (FL), and Mantle Cell Lymphoma (MCL). Currently, class-specific probes are used in order to reliably distinguish the sub-types, but these come with additional cost and equipment overheads. Expert pathologists specializing in these types of lymphomas, on the other hand, have shown promise in being able to differentiate these sub-types on H&E, indicating that there is the potential for a DP approach to be employed. A successful approach would allow for more consistent and less demanding diagnosis of this disease. This dataset was created to mirror real-world situations and as such contains samples prepared by different pathologists at different sites. They have additionally selected samples which contain a larger degree of staining variation than one would normally expect.
This use case represents the only classification use case of this manuscript: attempting to separate images into 1 of 3 sub-types of lymphoma. In the previous tasks, we were looking at primitives and attempting to segmented or detect them. In this case, though, a high level approach is taken, wherein we provide whole tissue samples to have the DL learn unique features of each class.
We break down this approach into 5 steps:
Step 1: Patch Extraction (Matlab): extract patches from all images separated into the 3 sub-types
Step 2: Cross-Validation Creation (Matlab): at the image level, split the patches into a 5-fold training and testing sets
Step 3: Database Creation (Bash): using the patches and training lists created in the previous steps, create 5 sets of leveldb training and testing databases, with mean files, for high performance DL training.
Step 4: Training of DL classifier (Bash): Slightly alter the 2 prototxt files used by Caffe, the solver and the architecture to point to the correct file locations. Use these to train the classifier.
Step 5: Generating Output on Test Images (Python): Use final model to generate the output
There are, of course, other ways of implementing a pipeline like this (e.g., use Matlab to directly create a leveldb, or skip the leveldb entirely, and use the images directly for training) . I’ve found using the above pipeline fits easiest into the tools that are available inside of Caffe and Matlab, and thus requires the less maintenance and reduces complexity for less experienced users. If you have a suggested improvement, I’d love to hear it!
The dataset consist of 374 images of size 1388 x 1040. These are further broken down into 113 for the CLL class, 139 for the FL class and 122 for the MCL class. Unfortunately, there is no description with the data indicating if the prefix of the file name indicates a unique patient or a unique facility. They did indicate that the data has been curated from multiple sources to create a real-world type cohort which contains typical stain and scanning variances.
Regardless, create a valid comparison to wnd-chrm, we treat the images in the same way and assume that each image is from a unique patient. which also used this dataset,
The data is located here (1.4G).
Examples of these images can be seen below
Step 1: Patch Extraction (Matlab)
We refer to step1_make_patches.m, which is fully commented.
A high level understanding is provided here:
- Since each image is representative of the entire class (as opposed to at a pixel level), we don’t have (or need) any annotation masks. So we simply break each image up into patches and assign them to the appropriate class.
- Each patch is saved to disk. At the same time, we maintain a “class_struct”, which contains all of the file names which have been written to disk.
u_sub_z.png — > example CLL-sj-03-476_001_sub_1.png
The file name format for each patch is as follows:
Where u is the image ID, (CLL-sj-03-476_001), which has the prefix of the class, followed by z which indicates what patch number it is. In this case, we haven’t kept track of the rotations, although there are two (0 and 90), thus odd numbers are 0 degree rotation (z=1,3,5,…) and even numbers are 90 degree rotation (z=2,4,6,….)
Step 2: Cross-Validation Creation (Matlab)
Now that we have all of the patches written to disk, and we have all of the file names saved into patient_struct, we want to split them into a cross fold validation scheme. We use step2_make_training_lists.m for this, which is fully commented.
In this code, we use a 5-fold validation, for each fold, we create 4 text files. Using fold 1 as an example:
train_w32_parent_1.txt: This contains a list of the patient IDs which have been used as part of this fold’s training set. This is similar to test_w32_parent_1.txt, which contains the patients used for the test set. An example of the file content is:
train_w32_1.txt: contains the filenames of the patches which should go into the training set (and test set when using test_w32_1.txt). The file format is [filename] [tab] [class]. Where class is 0,1,2 for CLL, FL and MCL, respectively. An example of the file content is:
All done with the Matlab component!
Step 3: Database Creation (Bash)
Now that we have both the patches saved to disk, and training and testing lists split into a 5-fold validation cohort, we need to get the data ready for consumption by Caffe. It is possible, at this point, to use an Image layer in Caffe and skip this step, but it comes with 2 caveats, (a) you need to make your own mean-file and ensure it is in the correct format and (b) an image layer can is not designed for high throughput. Also, having 100k+ files in a single directory can bring the system to its knees in many cases (for example, “ls”, “rm”, etc), so it’s a bit more handy to compress them all in to 10 databases (1 training and 1 testing for 5 folds), and use Caffe’s tool to compute the mean-file.
For this purpose, we use this bash file: step3_make_dbs.sh
We run it in the “subs” directory (“./” in these commands), which contains all of the patches. As well, we assume the training lists are in “../”, the directory above it.
Here we’ll briefly discuss the general idea of the commands, while the script has additional functionality (computes everything in parallel for example).
We use the caffe supplied convert_imageset tool to create the databases using this command:
~/caffe/build/tools/convert_imageset -shuffle -backend leveldb ./ DB_train_1
We first tell it that we want to shuffle the lists, this is very important. Our lists are in patient and class order, making them unsuitable for stochastic gradient descent. Since the database stores files, as supplied, sequentially, we need to permute the lists. Either we can do it manually (e.g., use sort –random) , or we can just let Caffe do it 🙂
We specify that we want to use a leveldb backend instead of a lmdb backend. My experiments have shown that leveldb can actually compress data much better without the consequence of a large amount of computational overhead, so we choose to use it.
Then we supply the directory with the patches, supply the training list, and tell it where to save the database. We do this similarly for the test set.
Creating mean file
To zero the data, we compute mean file, which is the mean value of a pixel as seen through all the patches of the training set. During training/testing time, this mean value is subtracted from the pixel to roughly “zero” the data, improving the efficiency of the DL algorithm.
Since we used a levelDB database to hold our patches, this is a straight forward process:
~/caffe/build/tools/compute_image_mean DB_train_1 DB_train_w32_1.binaryproto -backend leveldb
Supply it the name of the database to use, the mean filename to use as output and specify that we used a leveldb backend. That’s it!
Step 4: Training of DL classifier (Bash)
Now that we have the databases, and the associated mean-files, we can use Caffe to train a model.
There are two files which need to be slightly altered, as discussed below:
BASE-alexnet_solver.prototxt: This file describes various learning parameters (iterations, learning method (Adagrad) etc).
On lines 1 and 10 change: “%(kfoldi)d” to be the number of the fold for training (1,2,3,4,5).
On line 2: change “%(numiter)d” to number_test_samples/128. This is to have Caffe iterate through the entire test database. Its easy to figure out how many test samples there are using:
“wc –l test_w32_1.txt”
BASE-alexnet_traing_32w_db.prototxt: This file defines the architecture.
We only need to change lines 8, 12, 24, and 28 to point to the correct fold (again, replace “%(kfoldi)d” with the desired integer).
Also, in this use case, since we have 3 possible classes (CLL, FL, and MCL), we need to change line 173 from “num_output: 2” to “num_output: 3”
Note, these files assume that the prototxts are stored in a directory called ./model and that the DB files and mean files are stored in the directory above (../). You can of course use absolute file path names when in doubt.
In our case, we had access to a high performance computing cluster, so we used a python script (step4_submit_jobs.py) to submit all 5 folds to be trained at the same time. This script automatically does all of the above work, but you need to provide the working directory on line 11. I use this (BASE-qsub.pbs) PBS script to request resources from our Torque scheduler, which is easily adaptable to other HPC environments.
If you’ve used the HPC script above, things should already be queued for training. Otherwise, you can start the training simply by saying:
~/caffe/build/tools/caffe train –solver=1-alexnet_solver_ada.prototxt
In the directory which has the prototxt files. That’s it! Now wait until it finishes (600,000) iterations. 🙂
Step 5: Generating Output on Test Images (Python)
At this point, you should have a model available, to generate some output images. Don’t worry, if you don’t, you can use mine.
Here is a python script, to generate the test output for the associated k-fold (step5_generate_lymphoma_output.py).
It takes 1 command line arguments, the fold. In this case since we solely need to make a judgement for the entire image, we can compute a limited number of pixels, for example having a stride of 32 similar to our patch extraction technique.
The base directory is expected to contain:
BASE/images: a directory which contains the tif images for output generation
BASE/models: a directory which holds the 5 models (1 for each fold)
BASE/test_w32_parent_X.txt: the list of parent IDs to use in creating the output for fold X=1,2,3,4,5, created in step 2
BASE/DB_train_w32_X.binaryproto: the binary mean file for fold X=1,2,3,4,5, created in step 3
To compute the accuracy, and find out which images have been mis-classified, at a high level, this script:
- Determines the actual class of the image, based off of the first letter of the file name: C,F,M
- Extracts patches and run them through the classifier, obtain their predicted class (0,1,2 for CLL, FL and MCL respectively)
- Computes their overall frequency as “votes” per class
- Takes the argmax of the frequencies to determine the overall predicted class for the image
- Update the confusion matrix
Efficiency in Patch Generation
Writing a large number of small, individual files to a harddrive (even SSD) is likely going to take a very long time. Thus for Step 1 & Step 2, I typically employ a ram disk to drastically speed up the processes. Regardless, make sure Matlab does not have the output directory in its search path, otherwise it will likely crash (or come to a halt), while trying to update its internal list of available files.
As well, using a Matlab Pool (matlabpool open), opens numerous workers which also greatly speed up the operation and is recommended as well.
It is very important to use the model on images of the same magnification as the training magnification. This is to say, if your patches are extracted at 40x, then the test images need to be done at 40x as well.
Code is available here
Data is available here (1.4G)