Project Abstract

This project focuses on creating a lightweight convolutional neural network to identify Emphysema from lung x-rays. The motivation was that emphysema is rising due to environmental pollutants, smoking cigarettes and e-cigarettes with hospital resources not being able to keep up. Emphysema is identified easily through HRCT images; however, x-rays are more accessible and common for diagnosis but it is challenging due to the complexity of the disease and the nature of x-rays. There is limited research into lightweight CNNs identifying Emphysema through x-rays.Throughout the project, many models were researched for various diseases and Emphysema became the focus as the accuracy was the lowest compared to other diseases. Models were created with various architectures, some having few layers with large kernels, some having many layers with small kernels and combinations of both. Training and testing each model for 100 epochs took up to 8hrs per adjustment, hence fine tuning was a cumbersome task and the final model can still be improved; however, the results are promising.The project has improved on lightweight models. It cannot compete against large models with high hardware requirements as in the medical field, you want to use the model with the fewest false positives and false negatives. However, the main finding is that it is possible to get a good accuracy (Accuracy: 78%, Precision: 0.83, Recall: 0.77) with a lightweight model. They do have the potential to be useful in training staff and supporting diagnosis in hospitals with limited resources where a scan of an x-ray would point staff towards the right diagnosis. With further fine-tuning and small lung patch analysis, lightweight x-ray analysis models may be comparable to modern HRCT models as the project has beaten older HRCT models trained on small patches whereas this model was trained on full x-rays.

Keywords: Medical Convolutional Neural Networks, Lightweight X-Ray Neural Network, Emphysema Diagnosis

 Conference Details

Session: Poster Session B at Poster Stand 58

Location: Sir Stanley Clarke Auditorium at Wednesday 8th 09:00 – 12:30

Markers: Lu Zhang, Monika Seisenberger

Course: BSc Computer Science, 3rd Year

Future Plans: I have a job lined-up