Dataset Description: (Dataset can be downloaded from here: https://www.kaggle.com/cjinny/mura-v11
- MURA is a large dataset of bone X-Rays containing 14,863 musculoskeletal studies of the upper extremity.
- Each study contains one or more views (images) and is labeled as either normal or abnormal.
- Our dataset, MURA, contains 9,045 normal and 5,818 abnormal musculoskeletal radiographic studies of the upper extremity including the shoulder, humerus, elbow, forearm, wrist, hand, and finger.
- The training dataset consists of 14,863 studies(9045 normal, 5818 abnormal) whereas the validation dataset contains 207 studies. Total number of images is 40,561.
- Exploratory Data Analysis is in folder 'EDA'.
Task Description:
- MURA abnormality detection task is a binary classification task, where the input is an upper extremity radiograph and the expected output is a binary label y ∈ {0, 1} indicating whether the radiograph is normal or abnormal, respectively.
Models implemented: I have implemented two models: InceptionV2 model and Xception model using Tensorflow.
Important points for both models:
- Before feeding the images to the network, each image is normalized to have same mean and standard deviation as 0 and 1 respectively, scaled to 224 x 224 and augmented using following hyperparameters:
- rotation_range = 30
- brightness_range = [0.8,1.2]
- horizontal_flip = True - Used Adam optimizer with learning_rate = 0.0001 and default values of betas.
- Used binary_crossentropy loss function.
- Used dropout regularisation of rate = 0.3 for inceptionV2 model.
- Used l2 regularisation of lambda = 1e-6 for Xception model
- Number of epochs ran for :- Xception model : 35 and inceptionV2 model : 80
Results:
| Results | Xception |
|---|---|
| prediction error | 0.195 |
| Cohen kappa score | 0.604 |
| f1_score | 0.80 |
| Accuracy | 0.819 |