Pneumonia Detection using machine learning - with PyTorch

Last update: Jul 07, 2022

Overview

Pneumonia Detection

Pneumonia Detection using machine learning.

Training was done in colab:

DEMO:

Result (Confusion Matrix):

Data

I uploaded my dataset to kaggle I used a modified version of this dataset from kaggle. Instead of NORMAL and PNEUMONIA I split the PNEUMONIA dataset to BACTERIAL PNUEMONIA and VIRAL PNEUMONIA. This way the data is more evenly distributed and I can distinguish between viral and bacterial pneumonia. I also combined the validation dataset with the test dataset because the validation dataset only had 8 images per class.

This is the resulting distribution:

Processing and Augmentation

I resized the images to 150x150 and because some images already were grayscale I also transformed all the images to grayscale.

Additionaly I applied the following transformations/augmentations on the training data:

transforms.Resize((150, 150)),
transforms.Grayscale(),
transforms.ToTensor(),
transforms.RandomHorizontalFlip(),
transforms.RandomVerticalFlip(),
transforms.RandomRotation(45)

and those transformations on the test data:

transforms.Resize((150, 150)),
transforms.Grayscale(),
transforms.ToTensor(),

This is the resulting data:

I also used one-hot encoding for the labels!

Model

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1         [-1, 16, 148, 148]             160
              ReLU-2         [-1, 16, 148, 148]               0
       BatchNorm2d-3         [-1, 16, 148, 148]              32
            Conv2d-4         [-1, 16, 146, 146]           2,320
              ReLU-5         [-1, 16, 146, 146]               0
       BatchNorm2d-6         [-1, 16, 146, 146]              32
         MaxPool2d-7           [-1, 16, 73, 73]               0
            Conv2d-8           [-1, 32, 71, 71]           4,640
              ReLU-9           [-1, 32, 71, 71]               0
      BatchNorm2d-10           [-1, 32, 71, 71]              64
           Conv2d-11           [-1, 32, 69, 69]           9,248
             ReLU-12           [-1, 32, 69, 69]               0
      BatchNorm2d-13           [-1, 32, 69, 69]              64
        MaxPool2d-14           [-1, 32, 34, 34]               0
           Conv2d-15           [-1, 64, 32, 32]          18,496
             ReLU-16           [-1, 64, 32, 32]               0
      BatchNorm2d-17           [-1, 64, 32, 32]             128
           Conv2d-18           [-1, 64, 30, 30]          36,928
             ReLU-19           [-1, 64, 30, 30]               0
      BatchNorm2d-20           [-1, 64, 30, 30]             128
        MaxPool2d-21           [-1, 64, 15, 15]               0
           Conv2d-22          [-1, 128, 13, 13]          73,856
             ReLU-23          [-1, 128, 13, 13]               0
      BatchNorm2d-24          [-1, 128, 13, 13]             256
           Conv2d-25          [-1, 128, 11, 11]         147,584
             ReLU-26          [-1, 128, 11, 11]               0
      BatchNorm2d-27          [-1, 128, 11, 11]             256
        MaxPool2d-28            [-1, 128, 5, 5]               0
          Flatten-29                 [-1, 3200]               0
           Linear-30                 [-1, 4096]      13,111,296
             ReLU-31                 [-1, 4096]               0
          Dropout-32                 [-1, 4096]               0
           Linear-33                 [-1, 4096]      16,781,312
             ReLU-34                 [-1, 4096]               0
          Dropout-35                 [-1, 4096]               0
           Linear-36                    [-1, 3]          12,291
          Softmax-37                    [-1, 3]               0
================================================================
Total params: 30,199,091
Trainable params: 30,199,091
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.09
Forward/backward pass size (MB): 27.95
Params size (MB): 115.20
Estimated Total Size (MB): 143.24
----------------------------------------------------------------

Visualization using Streamlit

The webapp is not hosted because the model is too large. I'd have to host it on a server. This is just to visualize.

Pneumonia Detection using machine learning - with PyTorch

Related tags

Overview

Pneumonia Detection

Data

Processing and Augmentation

Model

Visualization using Streamlit

Owner

Wilhelm Berghammer

Demonstrational Session git repo for H SAF User Workshop (28/1)

Deep Learning for Natural Language Processing SS 2021 (TU Darmstadt)

PyTorch implementation of Octave Convolution with pre-trained Oct-ResNet and Oct-MobileNet models

SLIDE : In Defense of Smart Algorithms over Hardware Acceleration for Large-Scale Deep Learning Systems

Learning Representational Invariances for Data-Efficient Action Recognition

Combining Automatic Labelers and Expert Annotations for Accurate Radiology Report Labeling Using BERT

Python scripts form performing stereo depth estimation using the high res stereo model in PyTorch .

PyZebrascope - an open-source Python platform for brain-wide neural activity imaging in behaving zebrafish

Deep learning (neural network) based remote photoplethysmography: how to extract pulse signal from video using deep learning tools

Code accompanying the paper Shared Independent Component Analysis for Multi-subject Neuroimaging

TensorFlow implementation of "Variational Inference with Normalizing Flows"

The Incredible PyTorch: a curated list of tutorials, papers, projects, communities and more relating to PyTorch.

Data reduction pipeline for KOALA on the AAT.

Code Release for the paper "TriBERT: Full-body Human-centric Audio-visual Representation Learning for Visual Sound Separation"

CT Based COVID 19 Diagnose by Image Processing and Deep Learning

Robotics environments

Streamlit app demonstrating an image browser for the Udacity self-driving-car dataset with realtime object detection using YOLO.

Implementation of the paper All Labels Are Not Created Equal: Enhancing Semi-supervision via Label Grouping and Co-training

Here I will explain the flow to deploy your custom deep learning models on Ultra96V2.

[ACMMM 2021 Oral] Enhanced Invertible Encoding for Learned Image Compression