Majid Afshar

M.D. , M.S

Loyola University Medical Center

Assistant Professor

NLST (Learn more about this study)

NLST-520

Jun 24, 2019

Artificial intelligence for detection of obstructive airway disease and malignant pulmonary nodules

The differential diagnosis for pulmonary nodules (PN) is broad, and although the majority are ultimately found to be benign, invasive procedures such as biopsy or resection are often required to rule out malignancy. Currently, decisions on whether to biopsy or resect a PN are informed by guidelines, such as those published by the American College of Chest Physicians and the British Thoracic Society, validated malignancy risk prediction models, or both. These guidelines and risk prediction models assess a patient’s risk based on clinical and radiographic characteristics. However, despite currently available guidelines and prediction models, a significant number of patients with benign PNs still undergo invasive procedures. These interventions carry inherent risks, therefore, better identification of malignant PNs would have significant clinical and economic implications for both the patient and the health system.

This study aims to build off previously established models and to develop a new prediction model using natural language processing (NLP) and radiomics to improve the risk stratification of pulmonary nodules. The emerging field of radiomics, where imaging data is converted into mineable data, has shown promise as a data source and identifier of phenotypes that can be used for PN malignancy prediction model building. Our model will use both radiomics and NLP, a branch of computer science that uses linguistics to analyze regular speech such as clinician notes and documentation. Using NLP with radiomics will allow us to mine both the electronic medical record as well as radiologic images and to combine the predictors identified by these processes into a novel PN malignancy risk prediction model. Our goal is to build a novel malignancy risk prediction model that will help clinician’s better risk-stratify patients with PNs with the goal of reducing the number of patients requiring invasive procedures.

A. To build a novel PN malignancy risk prediction model using NLP and radiomics.

Hypothesis: NLP and radiomics can be used to build a PN malignancy risk prediction that can better risk-stratify patients with PN and decrease the need for invasive procedures.

B. To predict severity of obstructive airway disease using NLP and radiomics.

Hypothesis: NLP and radiomics can be used to differentiate low, moderate, and severe emphysema.