Sign In

Zerotogans Aptos2019 Blindnessdetection

Mon Apr 11 01:50:37 2022 +-----------------------------------------------------------------------------+ | NVIDIA-SMI 460.32.03 Driver Version: 460.32.03 CUDA Version: 11.2 | |-------------------------------+----------------------+----------------------+ | GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |===============================+======================+======================| | 0 Tesla P100-PCIE... Off | 00000000:00:04.0 Off | 0 | | N/A 34C P0 26W / 250W | 0MiB / 16280MiB | 0% Default | | | | N/A | +-------------------------------+----------------------+----------------------+ +-----------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=============================================================================| | No running processes found | +-----------------------------------------------------------------------------+


APTOS 2019 Blindness Detection

Diabetic Retinopathy(DR) can cause blindness. There are stages of the disease and its root cause is Diabetes.

The World Health Organization reports that more than 300 million people worldwide have diabetes (Wong et al. 2016).

The disease has been rising rapidly in developing countries.Regular testing and Early detection are important steps to prevent DR.

The test procedure requires a trained clinical expert to examine the fundus photographs of the patient's retina. This creates delays in diagnosis and treatment. This is especially relevant for developing countries, which often lack qualified medical staff to perform the diagnosis.

Machine Model can help in the detection of DR. It can speed up the efficiency and coverage of the screening programs.

In this project we will develop a convolutional neural network (CNN) for predicting the severity of the diabetic retinopathy based on the patient's retina photos.

This project is a part of the Deep Learning Certification Course-'Zero to GANs' on Jovian.

In 2019,a Kaggle competition was hosted by the Asia Pacific Tele-Ophthalmology Society (APTOS).

We are provided with a large set of retina images taken using fundus photography under a variety of imaging conditions.

A clinician has rated each image for the severity of diabetic retinopathy on a scale of 0 to 4:

0 - No DR

1 - Mild

2 - Moderate

3 - Severe

4 - Proliferative DR

Like any real-world data set, there are variety of issues in both the images and labels. Images contain artifacts, vary in size, are out of focus, underexposed, or overexposed. The images were gathered from multiple clinics using a variety of cameras over an extended period of time, which will introduce further variation.


As part of this Deep Learning Course, I have learnt about building deep learning models in Pytorch for a variety of datasets building models using Logistic Regression, Neural Networks, CNNs, Resnets, GANs and Transfer Learning.

As this project involves Image Classification and has small set of training data, we will be using the Transfer Learning on Pretrained Models like Resnet50 and efficientnet-b0 in Pytorch.

The steps we will follow are:

  1. Importing the libraries
  2. Downloading Dataset
  3. Data exploration
  4. Image transformations/Preprocessing
  5. Preparing the Dataset/DataLoaders
  6. Training the model

The data includes 3,662 labeled retina images of clinical patients and a test set with 1,928 images with unknown labels.

!pip install jovian --upgrade --quiet
import jovian
# Execute this to save new versions of the notebook
[jovian] Detected Colab notebook... [jovian] Please enter your API key ( from ): API KEY: ·········· [jovian] Uploading colab notebook to Jovian... Committed successfully!
Shubhda Datta6 months ago