Open in app

Sign in

Write

Sign in

Acute Lymphoblastic Leukemia detection with Tensorflow

AllDS2020 Classifier

Adam Milton-Barker
Peter Moss Leukaemia MedTech Research CIC

--

This project is the classifier that is used in Acute the Lymphoblastic Leukemia Detection System 2020. The network provided in this project was originally created in my ALL research papers evaluation project, where I replicated the network proposed in the Acute Leukemia Classification Using Convolution Neural Network In Clinical Decision Support System paper by Thanh.TTP, Giao N. Pham, Jin-Hyeok Park, Kwang-Seok Moon, Suk-Hwan Lee, and Ki-Ryong Kwon, and the data augmentation proposed in Leukemia Blood Cell Image Classification Using Convolutional Neural Network by T. T. P. Thanh, Caleb Vununu, Sukhrob Atoev, Suk-Hwan Lee, and Ki-Ryong Kwon. The original project was inspired by the work done by Amita Kapoor and Taru Jain and my previous projects based on their work.

Results

Network Architecture

Fig 1. Proposed architecture (Source)

In Acute Leukemia Classification Using Convolution Neural Network In Clinical Decision Support System, the authors propose a simple 5 layer Convolutional Neural Network.

In this project we will use an augmented dataset with the network proposed in this paper, built using Tensorflow 2.

We will build a Convolutional Neural Network, as shown in Fig 1, consisting of the following 5 layers (missing out the zero padding layers):

  • Conv layer (50x50x30)
  • Conv layer (50x50x30)
  • Max-Pooling layer (25x25x30)
  • Fully Connected layer (2 neurons)
  • Softmax layer (Output 2)

To get started make sure you completed the steps on the project home README.

Once you have your data you need to add it to the project filesystem. You will notice the data folder in the Model directory, Model/Data, inside you have ALL-IDB-1 & inside there you have Test.

If you are using data augmentation, which this tutorial assumes, first take ten positive and ten negative samples and place them in the Model/Data/ALL-IDB-1/Test directory. This will be used by our Oculus Rift application and our testing purposes. In my case I used:

  • im006_1.jpg
  • im020_1.jpg
  • im024_1.jpg
  • im026_1.jpg
  • im028_1.jpg
  • im031_1.jpg
  • im035_0.jpg
  • im041_0.jpg
  • im047_0.jpg
  • im053_1.jpg
  • im057_1.jpg
  • im060_1.jpg
  • im063_1.jpg
  • im069_0.jpg
  • im074_0.jpg
  • im088_0.jpg
  • im095_0.jpg
  • im099_0.jpg
  • im101_0.jpg
  • im106_0.jpg

Next add the remaining 88 images to the ALL-IDB-1 folder. The test images used in the demo will not be augmented, which I believe would be the case in a real world scenario.

The code for this project consists of 4 main Python files and a configuration file:

  • config.json: The configuration file.
  • AllDS2020.py: Core classifier wrapper class.
  • Helpers.py: A helper class.
  • Data.py: A data helpers class.
  • Model.py: A model helpers class.

Our functionality for this network can be found mainly in the Classes directory.

Helpers.py

Helpers.py is a helper class. The class loads the configuration and logging that the project uses.

Data.py

Data.py is a data helper class. The class provides the functionality for sorting and preparing your training and validation data.

Model.py

Model.py is a model helper class. The class provides the functionality for creating our CNN.

Configuration

config.json holds the configuration for our network.

{
    "cnn": {
        "api": {
            "server": "XXX.XXX.X.XXX",
            "port": 1234
        },
        "model": {
            "model": "Model/model.json",
            "model_aug": "Model/model_augmentation.json",
            "weights": "Model/weights.h5",
            "weights_aug": "Model/weights_augmentation.h5"
        },
        "data": {
            "dim": 50,
            "dim_augmentation": 100,
            "file_type": ".jpg",
            "labels": [0,1],
            "rotations_augmentation": 5,
            "seed_adam": 32,
            "seed_adam_augmentation": 32,
            "seed_rmsprop": 3,
            "seed_rmsprop_augmentation": 6,
            "split": 0.255,
            "split_augmentation": 0.3,
            "test": "Model/Data/ALL-IDB-1/Test",
            "train_dir": "Model/Data/ALL-IDB-1",
            "valid_types": [
              ".JPG",
              ".JPEG",
              ".PNG",
              ".GIF",
              ".jpg",
              ".jpeg",
              ".png",
              ".gif"
            ]
        },
        "train": {
            "batch": 80,
            "batch_augmentation": 100,
            "decay_adam": 1e-6,
            "decay_rmsprop": 1e-6,
            "epochs": 150,
            "epochs_augmentation": 150,
            "learning_rate_adam": 1e-4,
            "learning_rate_rmsprop": 1e-4,
            "val_steps": 10,
            "val_steps_augmentation": 3
        }
    }
}

We have the cnn object containing two objects, data and train. In data we have the configuration related to preparing the training and validation data. We use a seed to make sure our results are reproducible. In train we have the configuration related to training the model.

In my case, the configuration above was the best out of my testing, but you may find different configurations work better. Feel free to update these settings to your liking, and please let us know of your experiences.

We can use metrics to measure the effectiveness of our model. In this network we will use the following metrics:

tf.keras.metrics.BinaryAccuracy(name='accuracy'), tf.keras.metrics.Precision(name='precision'), tf.keras.metrics.Recall(name='recall'), tf.keras.metrics.AUC(name='auc')

These metrics will be displayed and plotted once our model is trained. A useful tutorial while working on the metrics was the Classification on imbalanced data tutorial on Tensorflow’s website.

Now you are ready to train your model. As mentioned above, I used an Ubuntu machine with an NVIDIA GTX 1050 ti, using different machines/GPU/CPU may vary the results, if so please let us know your findings.

Ensuring you have completed all previous steps, you can start training using the following command.

python3 AllDS2020.py Train Adam True

This tells the classifier to start in Train mode, use the Adam optimizer, and to to use data augmentation. You can play around and use rmsprop optimizer and no data augmentation etc, but for this tutorial the above is the command to use.

Data

First the data will be prepared.

2020-03-05 21:03:35,817 - Data - INFO - All data: 88
2020-03-05 21:03:35,817 - Data - INFO - Positive data: 39
2020-03-05 21:03:35,817 - Data - INFO - Negative data: 49
2020-03-05 21:03:40,155 - Data - INFO - Data shape: (1144, 100, 100, 3)
2020-03-05 21:03:40,157 - Data - INFO - Labels shape: (1144, 2)
2020-03-05 21:03:40,157 - Data - INFO - Augmented data: (1144, 100, 100, 3)
2020-03-05 21:03:40,157 - Data - INFO - All Labels: (1144, 2)
2020-03-05 21:03:40,684 - Data - INFO - Training data: (852, 100, 100, 3)
2020-03-05 21:03:40,684 - Data - INFO - Training labels: (852, 2)
2020-03-05 21:03:40,684 - Data - INFO - Validation data: (292, 100, 100, 3)
2020-03-05 21:03:40,684 - Data - INFO - Validation labels: (292, 2)

Model Summary

Our network matches the architecture proposed in the paper exactly, with exception to maybe the optimizer and loss function as this info was not provided in the paper.

Before the model begins training, we will be shown the model summary, or architecture.

Model: "AllCnn" _________________________________________________________________ Layer (type) Output Shape Param # ================================================================= zero_padding2d (ZeroPadding2 (None, 104, 104, 3) 0 _________________________________________________________________ conv2d (Conv2D) (None, 100, 100, 30) 2280 _________________________________________________________________ zero_padding2d_1 (ZeroPaddin (None, 104, 104, 30) 0 _________________________________________________________________ conv2d_1 (Conv2D) (None, 100, 100, 30) 22530 _________________________________________________________________ max_pooling2d (MaxPooling2D) (None, 50, 50, 30) 0 _________________________________________________________________ flatten (Flatten) (None, 75000) 0 _________________________________________________________________ dense (Dense) (None, 2) 150002 _________________________________________________________________ activation (Activation) (None, 2) 0 ================================================================= Total params: 174,812 Trainable params: 174,812 Non-trainable params: 0 Train on 852 samples, validate on 292 samples

Below are the training results for 150 epochs.

Fig 2. Ubuntu/GTX 1050 ti Accuracy

Fig 3. Ubuntu/GTX 1050 ti Loss

Fig 4. Ubuntu/GTX 1050 ti Precision

Fig 5. Ubuntu/GTX 1050 ti Recall

Fig 6. Ubuntu/GTX 1050 ti AUC

2020-03-05 21:09:05,268 - Model - INFO - Metrics: loss 0.10147356269370815
2020-03-05 21:09:05,268 - Model - INFO - Metrics: acc 0.9794521
2020-03-05 21:09:05,268 - Model - INFO - Metrics: precision 0.9794521
2020-03-05 21:09:05,268 - Model - INFO - Metrics: recall 0.9794521
2020-03-05 21:09:05,268 - Model - INFO - Metrics: auc 0.9939248

2020-03-05 21:09:12,757 - Model - INFO - Confusion Matrix: [[160   0]
 [  6 126]]

2020-03-05 21:09:15,365 - Model - INFO - True Positives: 126(43.15068493150685%)
2020-03-05 21:09:15,365 - Model - INFO - False Positives: 0(0.0%)
2020-03-05 21:09:15,365 - Model - INFO - True Negatives: 160(54.794520547945204%)
2020-03-05 21:09:15,365 - Model - INFO - False Negatives: 6(2.0547945205479454%)
2020-03-05 21:09:15,365 - Model - INFO - Specificity: 1.0
2020-03-05 21:09:15,365 - Model - INFO - Misclassification: 6(2.0547945205479454%)

Now we will use the test data to see how the classifier reacts to our testing data. Real world testing is the most important testing, as it allows you to see the how the model performs in a real world environment.

This part of the system will use the test data from the Model/Data/ALL-IDB-1/Test directory. The command to start testing locally is as follows:

python3 AllDS2020.py Classify Adam True

Results/Output

2020-03-05 20:56:17,227 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im060_1.jpg
2020-03-05 20:56:18,048 - Model - INFO - ALL correctly detected (True Positive)
2020-03-05 20:56:18,122 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im047_0.jpg
2020-03-05 20:56:18,125 - Model - INFO - ALL correctly not detected (True Negative)
2020-03-05 20:56:18,197 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im057_1.jpg
2020-03-05 20:56:18,201 - Model - INFO - ALL correctly detected (True Positive)
2020-03-05 20:56:18,230 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im026_1.jpg
2020-03-05 20:56:18,233 - Model - INFO - ALL correctly detected (True Positive)
2020-03-05 20:56:18,297 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im041_0.jpg
2020-03-05 20:56:18,301 - Model - INFO - ALL correctly not detected (True Negative)
2020-03-05 20:56:18,329 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im020_1.jpg
2020-03-05 20:56:18,332 - Model - INFO - ALL correctly detected (True Positive)
2020-03-05 20:56:18,396 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im053_1.jpg
2020-03-05 20:56:18,400 - Model - INFO - ALL correctly detected (True Positive)
2020-03-05 20:56:18,462 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im101_0.jpg
2020-03-05 20:56:18,466 - Model - INFO - ALL correctly not detected (True Negative)
2020-03-05 20:56:18,498 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im028_1.jpg
2020-03-05 20:56:18,502 - Model - INFO - ALL correctly detected (True Positive)
2020-03-05 20:56:18,566 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im069_0.jpg
2020-03-05 20:56:18,570 - Model - INFO - ALL correctly not detected (True Negative)
2020-03-05 20:56:18,629 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im095_0.jpg
2020-03-05 20:56:18,633 - Model - INFO - ALL incorrectly detected (False Positive)
2020-03-05 20:56:18,665 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im024_1.jpg
2020-03-05 20:56:18,668 - Model - INFO - ALL correctly detected (True Positive)
2020-03-05 20:56:18,736 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im088_0.jpg
2020-03-05 20:56:18,740 - Model - INFO - ALL incorrectly detected (False Positive)
2020-03-05 20:56:18,807 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im035_0.jpg
2020-03-05 20:56:18,811 - Model - INFO - ALL correctly not detected (True Negative)
2020-03-05 20:56:18,842 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im006_1.jpg
2020-03-05 20:56:18,845 - Model - INFO - ALL correctly detected (True Positive)
2020-03-05 20:56:18,910 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im074_0.jpg
2020-03-05 20:56:18,914 - Model - INFO - ALL correctly not detected (True Negative)
2020-03-05 20:56:18,975 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im099_0.jpg
2020-03-05 20:56:18,980 - Model - INFO - ALL correctly not detected (True Negative)
2020-03-05 20:56:19,045 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im063_1.jpg
2020-03-05 20:56:19,049 - Model - INFO - ALL correctly detected (True Positive)
2020-03-05 20:56:19,110 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im106_0.jpg
2020-03-05 20:56:19,114 - Model - INFO - ALL correctly not detected (True Negative)
2020-03-05 20:56:19,141 - Model - INFO - Loaded test image Model/Data/ALL-IDB-1/Test/Im031_1.jpg
2020-03-05 20:56:19,144 - Model - INFO - ALL incorrectly not detected (False Negative)
2020-03-05 20:56:19,145 - Model - INFO - Images Classifier: 20
2020-03-05 20:56:19,145 - Model - INFO - True Positives: 9
2020-03-05 20:56:19,145 - Model - INFO - False Positives: 2
2020-03-05 20:56:19,145 - Model - INFO - True Negatives: 8
2020-03-05 20:56:19,145 - Model - INFO - False Negatives: 1

Now we will use the test data to see how the server classifier reacts.

This part of the system will use the test data from the Model/Data/ALL-IDB-1/Test directory.

You need to open two terminal windows or tabs, in the first, use the following command to start the server:

python3 AllDS2020.py Server Adam True

In your second terminal, use the following command:

python3 AllDS2020.py Client Adam True

Results/Output

2020-03-05 23:44:00,128 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im060_1.jpg
2020-03-05 23:44:01,199 - Model - INFO - ALL correctly detected (True Positive)

2020-03-05 23:44:08,206 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im047_0.jpg
2020-03-05 23:44:08,555 - Model - INFO - ALL correctly not detected (True Negative)

2020-03-05 23:44:15,562 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im057_1.jpg
2020-03-05 23:44:15,908 - Model - INFO - ALL correctly detected (True Positive)

2020-03-05 23:44:22,910 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im026_1.jpg
2020-03-05 23:44:23,067 - Model - INFO - ALL correctly detected (True Positive)

2020-03-05 23:44:30,074 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im041_0.jpg
2020-03-05 23:44:30,405 - Model - INFO - ALL correctly not detected (True Negative)

2020-03-05 23:44:37,411 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im020_1.jpg
2020-03-05 23:44:37,602 - Model - INFO - ALL correctly detected (True Positive)

2020-03-05 23:44:44,610 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im053_1.jpg
2020-03-05 23:44:44,958 - Model - INFO - ALL correctly detected (True Positive)

2020-03-05 23:44:51,965 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im101_0.jpg
2020-03-05 23:44:52,339 - Model - INFO - ALL correctly not detected (True Negative)

2020-03-05 23:44:59,346 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im028_1.jpg
2020-03-05 23:44:59,534 - Model - INFO - ALL correctly detected (True Positive)

2020-03-05 23:45:06,538 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im069_0.jpg
2020-03-05 23:45:06,898 - Model - INFO - ALL correctly not detected (True Negative)

2020-03-05 23:45:13,902 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im095_0.jpg
2020-03-05 23:45:14,260 - Model - INFO - ALL incorrectly detected (False Positive)

2020-03-05 23:45:21,266 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im024_1.jpg
2020-03-05 23:45:21,427 - Model - INFO - ALL correctly detected (True Positive)

2020-03-05 23:45:28,435 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im088_0.jpg
2020-03-05 23:45:28,815 - Model - INFO - ALL incorrectly detected (False Positive)

2020-03-05 23:45:35,822 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im035_0.jpg
2020-03-05 23:45:36,165 - Model - INFO - ALL correctly not detected (True Negative)

2020-03-05 23:45:43,173 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im006_1.jpg
2020-03-05 23:45:43,336 - Model - INFO - ALL correctly detected (True Positive)

2020-03-05 23:45:50,342 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im074_0.jpg
2020-03-05 23:45:50,673 - Model - INFO - ALL incorrectly detected (False Positive)

2020-03-05 23:45:57,678 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im099_0.jpg
2020-03-05 23:45:58,062 - Model - INFO - ALL correctly not detected (True Negative)

2020-03-05 23:46:05,066 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im063_1.jpg
2020-03-05 23:46:05,401 - Model - INFO - ALL correctly detected (True Positive)

2020-03-05 23:46:12,404 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im106_0.jpg
2020-03-05 23:46:12,734 - Model - INFO - ALL correctly not detected (True Negative)

2020-03-05 23:46:19,738 - Model - INFO - Sending request for: Model/Data/ALL-IDB-1/Test/Im031_1.jpg
2020-03-05 23:46:19,932 - Model - INFO - ALL incorrectly not detected (False Negative)

2020-03-05 23:46:26,938 - Model - INFO - Images Classifier: 20
2020-03-05 23:46:26,938 - Model - INFO - True Positives: 9
2020-03-05 23:46:26,939 - Model - INFO - False Positives: 3
2020-03-05 23:46:26,939 - Model - INFO - True Negatives: 7
2020-03-05 23:46:26,939 - Model - INFO - False Negatives: 1

The Peter Moss Acute Myeloid & Lymphoblastic Leukemia AI Research project encourages and welcomes code contributions, bug fixes and enhancements from the Github.

Please read the CONTRIBUTING document for a full guide to forking our repositories and submitting your pull requests. You will also find information about our code of conduct on this page.

We use SemVer for versioning. For the versions available, see Releases.

This project is licensed under the MIT License — see the LICENSE file for details.

We use the repo issues to track bugs and general requests related to using this project. See CONTRIBUTING for more info on how to submit bugs, feature requests and proposals.

Originally published at https://github.com.

Artificial Intelligence
Artificial Neural Network
Leukemia
Research

--

--

Adam Milton-Barker
Peter Moss Leukaemia MedTech Research CIC

Written by Adam Milton-Barker

11 Followers
Editor for

Founder at Peter Moss Leukemia AI Research #PeterMossLeukemiaAiResearch #IntelSoftwareInnovator

Help

Status

About

Careers

Press

Blog

Privacy

Terms

Text to speech

Teams