2022 World Conference on Lung Cancer (ePosters)-EP01.05-007. Radiomics Based Machine Learning Model for Sub-cm Lung Nodule Malignancy Diagnosis in the PanCan Screening Study

EP01.05-007. Radiomics Based Machine Learning Model for Sub-cm Lung Nodule Malignancy Diagnosis in the PanCan Screening Study

Back to course

Pdf Summary

A study conducted by researchers at the BC Cancer Research Institute in Canada aimed to develop a machine learning model that can predict the malignancy risk of sub-cm lung nodules using low-dose CT (LDCT) images and radiomics. LDCT screening programs often identify small nodules that are difficult for physicians to classify as malignant due to the limited amount of pixel-based information. These small nodules are typically assigned a follow-up scan instead of immediate treatment.<br /><br />The study used 633 nodule CT cases from the PanCan screening trial, consisting of 456 patients. Of these cases, 142 were confirmed to be cancerous and 491 were benign based on a 5-year follow-up. Each nodule was segmented and features were extracted from LDCT images. These images were used to train a machine learning model for nodule malignancy diagnosis.<br /><br />The results showed that the machine learning model had a sensitivity of 85.0% and specificity of 77.3% for all nodules. For sub-cm nodules, the sensitivity was 80.7% and the specificity was 78.9%. When only the area feature was considered, the sensitivity was 83.3% and the specificity was 59.9%. The model also performed reasonably well on the PanCan Nodule Malignancy Risk Prediction Calculator (NMPC) subset, with a sensitivity of 58.6% and specificity of 79.8%.<br /><br />The researchers concluded that using artificial intelligence methods, they can improve the management strategy for small lung nodules, specifically for those assigned a "wait and see" follow-up CT scan. They also suggested future work to implement radiomics in combination with CT images for multi-input deep learning models and external validation.<br /><br />Overall, the study demonstrates the potential of machine learning and radiomics in improving the diagnosis and management of sub-cm lung nodules detected through LDCT screening.

Asset Subtitle

Ian Janzen

Meta Tag

Speaker Ian Janzen

Topic Early Detection and Screening - Pulmonary Nodule

Keywords

machine learning

malignancy risk

sub-cm lung nodules

low-dose CT

LDCT images

radiomics

nodule CT cases

PanCan screening trial

artificial intelligence methods

diagnosis and management