LUNG CANCER DETECTION IN LOW-RESOLUTION IMAGES

Abstract

One of the most important prognostic factors for all lung cancer patients is the accurate detection of metastases. Pathologists, as we all know, examine the body and its tissues. On the existing clinical method, they have a tedious and manual task. Recent analysis has been inspired by these aspects. Deep Learning (DL) algorithms have been used to identify lung cancer. The developed cutting-edge technologies beat pathologists in terms of cancer identification and localization inside pathology images. These technologies, though, are not medically feasible because they need a massive amount of time or computing capabilities to perceive high-resolution images. Image processing techniques are primarily employed for lung cancer prediction and early identification and therapy to avoid lung cancer. This research aimed to assess lung cancer diagnosis by employing DL algorithms and low-resolution images. The goal would be to see if Machine Learning (ML) models might be created that generate higher confidence conclusions while consuming fractional resources by comparing low and high-resolution images. A DL pipeline has been built to a small enough size from compressing high-resolution images to be fed into an or before CNN (Convolutional Neural Network) for binary classification i.e. cancer or normal. Numerous enhancements have been done to increase overall performance, providing data augmentations, including augmenting training data and implementing tissue detection. Finally, the created low-resolution models are practically incapable of handling extremely low-resolution inputs i.e. 299 x 299 to 2048 x 2048 pixels. Considering the lack of classification ability, a substantial reduction in models’ predictable times is only a marginal benefit. Due to an obvious drawback with the methodology, this is disheartening but predicted finding: very low resolutions, essentially expanding out on a slide, preserve only data about macro-cellular structures, which is usually insufficient to diagnose cancer by itself.

Keywords

• convolutional neural networks
• deep learning
• image processing
• lung cancer detection
• machine learning
• medical image analysis