Correlation of clinical information with imaging values gained from chest CTs analyzed with deep learning algorithm
The research will provide robust tools for correlating clinical data with quantitative lung tissue analysis using the CAD software contextflow ADVANCE Chest CT, Vienna/Austria enabling better predictions of disease progression and treatment response. The findings will support the development of personalized treatment strategies and enhance the application of imaging data in clinical settings.
Tissue characterization holds significant clinical value in diagnosing and managing various conditions, especially through advanced imaging techniques. While abdominal CT scans were traditionally used, chest CT offers the opportunity to extract more detailed and specific tissue information in the lung screening population. The complexity of lung patterns and the need for comprehensive analysis across large datasets underscore the potential benefits of leveraging artificial intelligence (AI) in this field.
By focusing on lung tissue quantification with CAD software, our project seeks to explore the correlation between clinical information from a lung screening dataset and lung tissue values obtained from these quantitative analyses. Understanding these relationships can be crucial for assessing the overall health of patients and identifying early indicators of conditions such as lung disease or cancer. Additionally, this analysis can provide insights into patient outcomes following weight-reduction therapies and help assess cancer patients' frailty before undergoing treatments such as chemotherapy or radiation.
We will aim to validate deep-learning models that use chest CT data for lung tissue characterization, offering insights into how clinical factors influence lung health. This will ultimately contribute to more personalized and data-driven approaches in radiology and clinical decision-making.
Our project has three primary aims to explore the relationship between clinical data and lung tissue characteristics in CT imaging in lung cancer screening population:
Aim 1: Using NLST data, we will refine and validate deep-learning models that quantify lung tissue characteristics, such as volume, density, and texture, based on chest CT scans in screening population. We will validate the models across clinical variables such as age, sex, and smoking history to ensure their accuracy and generalizability in clinical practice.
Aim 2:We will analyze the NLST dataset to investigate the relationship between clinical information (demographics, comorbidities, disease history, and treatment protocols) and lung tissue values obtained through CAD software. This step will involve statistical analyses to identify significant correlations between clinical parameters and quantified lung tissue characteristics, offering insights into lung disease progression, frailty, and recovery.
Aim 3: Using serial chest CT scans from the NLST dataset, we will evaluate how lung tissue characteristics change over time and how these changes relate to lung nodules and clinical outcomes, such as disease progression or recovery. Our models will track these changes and identify patterns that could inform future clinical interventions when it comes to lung cancer screening, detection, treatment and management.
Aim 4: Assess the clinical impact of the of AI-based lung pattern detection and quantification to enhance early diagnosis, personalized treatment strategies, and patient care management in lung cancer.
Johannes Hofmanninger, RnD Department Lead
Luc Nicodème, machine learning expert
Ester Jimenez, data scientist
Seastian Röhrich, M.D., radiologist