The Role of Radiogenomics in Lung Cancer Prognosis

Lung cancer remains one of the most prevalent forms of cancer globally and is the leading cause of cancer-related mortality. Despite advancements in medical research, the overall survival (OS) rate for lung cancer patients continues to be low, underlining the urgent need for more accurate diagnostic and prognostic tools. When used separately, traditional imaging and genomic data provide valuable but limited insights. Integrating these fields—known as radiogenomics—has emerged as a promising approach to enhance predictive accuracy in clinical outcomes. A recent review published by Insights into Imaging evaluates the efficacy of combining radiomic and genomic data in prognostic models for lung cancer.

Radiomics and Genomics in Prognostic Models

Radiomics uses high-throughput algorithms to extract large amounts of quantitative features from medical images, such as CT scans. These features can represent tumour shape, texture and intensity, allowing non-invasive predictions of tumour phenotype and behaviour. While radiomics has shown promise in predicting biomarkers and treatment responses, its standalone predictive power often faces challenges related to variability and reliability.

Genomics, on the other hand, involves analysing genetic variations and expressions that influence cancer progression and treatment outcomes. Genomic studies have revealed key genetic markers that correlate with tumour aggressiveness and patient survival rates. By itself, genomic analysis provides a biological basis for understanding cancer but may lack the spatial and structural context offered by imaging data.

Research has demonstrated that combining these two methodologies can result in a more comprehensive understanding of tumour characteristics. Integrating radiomic and genomic data into prognostic models allows for improved prediction of outcomes, such as overall survival, disease-free survival (DFS) and progression-free survival (PFS). Studies reviewed between 2016 and 2023 have shown that combined models can achieve area under the receiver operating characteristic (AUC) values as high as 0.99, surpassing models based on either data type alone. This improvement is particularly significant in clinical practice, where accurate prognosis informs treatment decisions and patient management.

Performance and Limitations of Combined Models

The performance of radiogenomic models has been a focal point in assessing their clinical utility. Out of the ten studies analysed, most showed that combination models outperformed those relying solely on radiomics or genomics. For example, one study achieved an AUC of 0.95 during training and 0.99 during validation when using a combination model. The C-index, another key performance metric, also indicated high predictive power in models that integrated both data sources, often ranging from 0.70 to 0.85. This suggests that the synergy between imaging features and genetic data provides a more robust framework for prognosis.

Despite these positive outcomes, limitations persist. Many studies included in the review had small sample sizes, ranging from 79 to 315 patients, which may not fully represent broader patient demographics. Additionally, most studies were retrospective in design, which could introduce selection biases and limit the applicability of the findings to general clinical practice. Variability in data acquisition methods, such as differences in CT imaging protocols and genetic sequencing techniques, further complicates the standardisation and reproducibility of these models. Ensuring consistency in data collection and model validation is essential for translating research findings into clinical workflows.

Another challenge lies in the methods used for feature selection and model construction. The least absolute shrinkage and selection operator (LASSO) method was the most commonly employed technique for selecting relevant radiomic and genomic features. While LASSO and multivariate Cox proportional hazard models proved effective for model construction, future research should explore alternative approaches that could offer better results in specific contexts. Although less commonly used, deep learning methods have shown potential for boosting model performance but come with the risk of overfitting, particularly in studies with limited data.

Methodological Insights and Future Directions

Developing a combined radiogenomic model involves multiple complex steps, from data acquisition and feature extraction to model training and validation. Radiomic feature extraction typically involves segmenting regions of interest (ROIs) from CT images and analysing them using specialised software. Genomic data, on the other hand, can be derived from various sources, such as tumour biopsy samples, and often require advanced bioinformatics tools for processing and integration with imaging data.

The choice of algorithms and statistical methods is critical in model performance. Multivariate Cox proportional hazard analysis was frequently used across the reviewed studies to identify predictive factors due to its computational efficiency and ability to handle multiple variables. However, some studies employed more complex machine learning techniques, such as support vector machines (SVMs) and convolutional neural networks (CNNs), to enhance prediction accuracy. A notable finding was that models combining traditional statistical approaches with machine learning often achieved the highest levels of accuracy.

Future research should aim to address the limitations of current studies by focusing on larger, multicentre trials with diverse patient populations. This will help validate findings and improve the robustness of the models. Standardisation of imaging protocols and genomic data acquisition methods is also crucial to ensure the reproducibility of results across different medical institutions. Integrating clinical data alongside radiomic and genomic information may offer even more accurate prognostic models. Such a multimodal approach could incorporate factors like patient history, treatment response, and lifestyle to further personalise lung cancer care.

The integration of radiomics and genomics in prognostic models for lung cancer offers significant potential for improving the accuracy of outcome predictions. Combined models have consistently shown better performance metrics compared to those based on radiomics or genomics alone. However, challenges such as small sample sizes, retrospective study designs, and variability in data acquisition need to be addressed for broader clinical application. Future research should emphasise prospective, multicentre studies and the development of standardised protocols to validate and enhance the practical use of these models. Ultimately, incorporating radiogenomic models into routine clinical practice could revolutionise the way lung cancer prognosis and treatment strategies are approached, providing a new layer of personalised care for patients.

Source: Insights into Imaging

Image Credit: iStock