PRE-THERAPY LUNG CANCER PROGNOSTIC PREDICTION IN IMAGES

Principal Investigator

Name
Stelmo Magalhaes Barros Neto

Degrees
Ph.D

Institution
Universidade Federal do Maranhão

Position Title
Professor

Email
snobnet@gmail.com

About this CDAS Project

Study
NLST (Learn more about this study)

Project ID
NLST-558

Initial CDAS Request Approval
Sep 5, 2019

Title
PRE-THERAPY LUNG CANCER PROGNOSTIC PREDICTION IN IMAGES

Summary
Machine learning methods, such as deep learning methods, can be used to generate images of possible lung injury locations. For this reason, as these methods use generative or classification-based adversearial neural networks and classification, they can be employed to predict the presence of lung cancer. Similarly, machine learning methods are used with texture characteristics to predict the diagnosis of lung cancer from an initial image (BARROS NETTO et al., 2019). Therefore, machine learning methods are useful tools in predicting lung cancer diagnosis not easily detected from an initial image.

Aims

The present work plan aims to predict the diagnosis of lung cancer from an initial image where lung cancer is not completely visible or localizable. This will be achieved through the use of machine learning methods, artificial intelligence and texture in images. Thus, from an initial image, a trained machine learning method will be able to diagnose lung cancer. Consequently, the specialist doctor can use the result obtained to choose whether to use more effort in cancer cases.
To achieve the overall goal of this work plan, some specific goals need to be achieved. The first specific objective of this work plan is to obtain a lung cancer temporal imaging database with information on lung cancer diagnosis and location. When obtaining this information, find a machine learning method and / or artificial intelligence and texture that uses this information to predict the presence or absence of lung cancer. Thus, the proposed method should produce the cancer diagnosis as accurately as possible and confirmed by images acquired later in subsequent years. Therefore, in case a convolutional deep learning method is not used, one more step should be employed, which will be the extraction of texture characteristics to feed a suitable machine learning method. Consequently, performance measures will be used to evaluate the rating obtained. Therefore, once these specific objectives have been achieved, our overall objective will have been raised.

Collaborators

Rhaylson Ribeiro Moreira