In this task, the objective is to develop a deep learning model capable of distinguishing between normal and abnormal X-ray images within medical studies. This task entails utilizing at least two distinct neural network architectures: one based on a custom Convolutional Neural Network (CNN) model and another employing a popular pre-trained CNN architecture, such as VGG-19 or ResNet. The models will be trained and evaluated using the MURA dataset, a comprehensive dataset of X-ray images from the Stanford ML group.
Dataset: The MURA dataset, provided by the Stanford ML group, is a collection of X-ray images that contain studies from various anatomical regions, including upper extremity, lower extremity, hand, wrist, elbow, and shoulder. Each study is labeled as either "normal" or "abnormal," and the dataset serves as the basis for training and evaluating the deep learning models.
Architectures:
-
Custom CNN Model: You will design and implement a custom CNN architecture tailored to the task. This allows for a more fine-tuned approach, specific to the characteristics of the medical images in the MURA dataset.
-
Pre-trained CNN Model: In addition to the custom CNN, you will utilize a popular pre-trained CNN architecture, such as VGG-19 or ResNet, for transfer learning. This leverages the knowledge learned from vast image datasets and can potentially enhance the model's performance.
Development Environment: The code for this project is implemented in Google Colab, providing a cloud-based Python environment that is conducive to deep learning tasks. The code and detailed implementation steps can be found within the accompanying PDF document, which contains links to the relevant Colab notebooks and additional resources for reference.
Project Summary: The primary objective of this task is to create a robust and accurate deep learning model for the automatic classification of X-ray studies into "normal" or "abnormal" categories. Leveraging both custom and pre-trained CNN architectures, you will demonstrate proficiency in developing models capable of aiding in the diagnosis of medical conditions using X-ray images. The code implementation and results will be available in the provided PDF document.