2024 World Conference on Lung Cancer (WCLC) - Posters-P4.07E.04 Pathological Grading of Early-Stage Lung Cancers Through CT-to-PET Translation and Co-Learning Model

Pdf Summary

The study outlined in the document focuses on an innovative model to improve the early detection and pathological grading of lung cancer using chest CT scans and synthetic PET imagery. With the expanding use of chest CT screening, there are increased detection rates of early-stage lung cancers, but challenges such as high false-positive rates and overdiagnosis persist. PET scans, although effective in categorizing nodules by pathology, pose issues due to their high cost, use of radioactivity, and technical complexities.<br /><br />The study introduces a co-learning model that generates PET images from CT scans, offering a refined approach to pathological grading. This model aims to improve the accuracy and cost-effectiveness of evaluating early-stage lung cancer, potentially enhancing early treatment decisions.<br /><br />A retrospective study conducted at Guangdong Provincial People's Hospital involved 642 patients with lung nodules sized between 4 mm and 30 mm, confirmed as benign or adenocarcinoma through surgical pathology. The study's model, referred to as the CPT (CT-to-PET translation) model, demonstrated excellent performance, achieving a high Structural Similarity Index Measure (SSIM) score of 0.956 and a low Root Mean Square Error (RMSE) of 0.0147. Radiologists in a Turing test rated 47% of the synthetic PET images as equivalent to ground-truth images.<br /><br />For pathological classification, the combined CT and synthetic PET model outperformed individual CT and PET models, showing an Area Under the Curve (AUC) of 0.941 for Cohort A and 0.869 for Cohort B. This indicates the model's enhanced predictive capability for early-stage pathology classification.<br /><br />Future research is suggested to include multicenter prospective samples for improved validation and adaptations of more advanced image translation models, ensuring further clinical applicability and broader utility in medical imaging advancement.

Asset Subtitle

Haiyu Zhou

Meta Tag

Speaker Haiyu Zhou

Topic Early-Stage NSCLC

Keywords

lung cancer

chest CT scans

synthetic PET imagery

early detection

pathological grading

co-learning model

CPT model

structural similarity index

pathological classification

medical imaging