Lung Cancer Screening - A non-invasive methodology for early diagnosis
Principal Investigator
Name
Hélder Oliveira
Degrees
Ph.D
Institution
INESC TEC - Institute for Systems and Computer Engineering, Technology and Science
Position Title
Research
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-615
Initial CDAS Request Approval
Dec 13, 2019
Title
Lung Cancer Screening - A non-invasive methodology for early diagnosis
Summary
Lung cancer is the most frequent, deadly and expensive cancer type, with an incidence of 1.8 million and mortality of 1.6 million worldwide in 2012. The main contributing factor to a successful treatment is an early diagnosis. With the current diagnosis model, radiologists must seek thoroughly for lung nodules in computerized tomography (CT).
This is a time consuming and often physically demanding procedure, that leads to errors, possible omissions and, consequently, late diagnosis. When nodules are found, most of the times they are sampled and biopsied to determine their malignancy and pathology. As such, there is an urgent need for a streamlined process able to allow diagnosis with greater efficiency and accuracy.
Computer-Aided Diagnosis (CAD) systems for automatic detection of pulmonary nodules in CT scans are one of the most studied CAD applications. Although these systems improve the overall performance of radiologists, current solutions often only allow the visual description of the tumours and lesions and thus are limited to a subjective and qualitative characterization.
With this project, we aim to push CAD systems to a more objective, both qualitative and quantitative tumour characterization, i.e., a radiomic approach to describe and to create predictive models relating images' phenotypes to genomics' signatures. We will develop a workable prototype to detection, classification and prognosis of lung lesions in chest CT scans. The models will be trained for characterization of lung cancer-associated lesions, using an annotated CT scans dataset, clinical and biological data collected retrospectively, provided by our clinical partners.
The predictive capability will be validated by state-of-the-art clinical and laboratorial studies. The development of the models will take advantage of the learning capabilities of machine learning approaches. The project will also have a prospective component, where a model will be developed to evaluate contributions of liquid biopsy in lung cancer characterization. In case of success, liquid biopsy will be of great value as a means to obtain molecular data in a minimally invasive way, compatible with the clinical routine.
The primary endpoint will be the creation of a private lung cancer image database with clinical and biological data to support radiomic approach. The secondary endpoint, will be the creation of predictive models relating images phenotypes to genomics' signatures, based on radiomics information and its correlation with biological data. Finally the third outcome will be the evaluation of the liquid biopsy as an early diagnostic tool. This proposal will have three levels of impact: increased sensitivity in the detection of lung lesions; improved CAD output (tumour characterization) - possibility of clinical stratification of lesions that are not eligible for tissue biopsy; increased specificity of the imaging based classification, with a potential decrease in the need for tissue biopsy.
This is a time consuming and often physically demanding procedure, that leads to errors, possible omissions and, consequently, late diagnosis. When nodules are found, most of the times they are sampled and biopsied to determine their malignancy and pathology. As such, there is an urgent need for a streamlined process able to allow diagnosis with greater efficiency and accuracy.
Computer-Aided Diagnosis (CAD) systems for automatic detection of pulmonary nodules in CT scans are one of the most studied CAD applications. Although these systems improve the overall performance of radiologists, current solutions often only allow the visual description of the tumours and lesions and thus are limited to a subjective and qualitative characterization.
With this project, we aim to push CAD systems to a more objective, both qualitative and quantitative tumour characterization, i.e., a radiomic approach to describe and to create predictive models relating images' phenotypes to genomics' signatures. We will develop a workable prototype to detection, classification and prognosis of lung lesions in chest CT scans. The models will be trained for characterization of lung cancer-associated lesions, using an annotated CT scans dataset, clinical and biological data collected retrospectively, provided by our clinical partners.
The predictive capability will be validated by state-of-the-art clinical and laboratorial studies. The development of the models will take advantage of the learning capabilities of machine learning approaches. The project will also have a prospective component, where a model will be developed to evaluate contributions of liquid biopsy in lung cancer characterization. In case of success, liquid biopsy will be of great value as a means to obtain molecular data in a minimally invasive way, compatible with the clinical routine.
The primary endpoint will be the creation of a private lung cancer image database with clinical and biological data to support radiomic approach. The secondary endpoint, will be the creation of predictive models relating images phenotypes to genomics' signatures, based on radiomics information and its correlation with biological data. Finally the third outcome will be the evaluation of the liquid biopsy as an early diagnostic tool. This proposal will have three levels of impact: increased sensitivity in the detection of lung lesions; improved CAD output (tumour characterization) - possibility of clinical stratification of lesions that are not eligible for tissue biopsy; increased specificity of the imaging based classification, with a potential decrease in the need for tissue biopsy.
Aims
This project will comprise 2 main objectives:
- describe and to create predictive models relating images phenotypes to genomics signatures and is pushing CAD systems to a more objective and quantitative tumour characterization based on retrospective (radiomic approach);
- develop predictive models for relating to evaluate contributions liquid biopsy in lung cancer characterization.
Collaborators
IPATIMUP-INSTITUTO DE PATOLOGIA E IMUNOLOGIA MOLECULAR DA UNIVERSIDADE DO PORTO, Portugal
Centro Hospitalar de São João Porto, Portugal