Pneumonia detection on chest X-ray exams

• Author fullname: Gustavo Ayrton Bitocchi
• Thesis' director: Diego Alexis Evin
• Austral University 2020/21 Cohort
• Master's final project

Requirements

Training and testing datasets.

Runtime environment

Pytorch framework was used for the creation and training of neural networks in the following sw environment:

• Python 3.10.4 version
• Pytorch 1.11.0 version
• CUDA 11.6.2 version
• Ubuntu 20.04 LTS

Directories

The 'Arquitectura-CNN-Autogenerada-CGP' directory contains the 3 following files:

• Generacion_Arquitectura_CNN_CGP_AWS_SageMaker.ipynb : Notebook that makes use of the CGP technique in order to automatically creates a CNN architecture.
• Entrenamiento_Arquitectura_Generada_CNN_CGP_AWS_SageMaker.ipynb : Notebook that trains the autogenerated CNN architecture.
• Evaluacion_Arquitectura_CNN_CGP.ipynb : Notebook that tests the autogenerated CNN architecture.

The 'Benchmark-ResNet50' directory contains the following file:

• Benchmark_ResNet50_AWS_SageMaker.ipynb: Notebook that performs Exploratory Data Analysis (EDA) over the training and testing datasets and creates a ResNet50 benchmark model.

The 'Grad-CAM' directory contains the following file:

• Grad_CAM_Modelo_ResNet50.ipynb: Notebook that uses a test image as an input and returns a heapmap.