Classifying Carcinomas in Patient Noisy Images with Lung Cancer
Josue Nataren
jdn36@duke.edu
| Paper PDF |
|
Lung Cancer is a very serious disease that causes the death of tens of thousands of people every year in the US and in the world. For this reason, there exists the need for better diagnostic tools to identify if a person has lung cancer, and if they do, what type to determine the severity of it. In this project, a neural network model is proposed to be able to classify CT-scan images into four different types of cancer cells: Adenocarcinoma, Squamous cell carcinoma, Large cell carcinoma, and Small cell carcinoma. Sometimes medical images can have noise in them; this is why a denoiser has been incorporated to the network so in this way images are cleaned and classified. The network showed potential in classification but needs more work in the denoiser component. It classified images at a 92% accuracy level, but the denoiser only had around 21% accuracy in cleaning the images. For future work, an improvement in the denoiser model needs to happen. With tools like this, patient-specific healthcare can be improved greatly.
|
|
|
| Paper: |
Code and Data:
|