Using CT imaging to find radiomic signatures in surgically resected NSCLC
Giulia Mazzaschi - University Hospital Parma, Parma, Italy
The presentation is about integrating tumour immune microenvironment parameters analysed at tissue level with the CT imaging features and specifically high throughput CT imaging features in terms of radiomics. Because we strongly believe that only through an integration of both of these aspects we can reach, or we can try to reach, a real precision oncology and a real personalised medicine. So this was the aim of our study, to some extent.
We started from groups of sixty surgically resected non-small cell lung cancer patients that we analysed, deeply analysed, in terms of tissue immune parameters, so PD-L1 expression, tumour infiltrating lymphocytes and whatever. From each CT scan of these patients performed before surgery we extracted specific radiomic features, specifically 841 radiomic features, so a huge amount of data, difficult to interpret but we are trying to do this. To do so we performed at first an unsupervised analysis in order to analyse, or to gross analyse, the distribution of these radiomic features among our patients. We identified specific classes of patients exhibiting a unique pattern of radiomic expression in this sense. Then we deeper analysed the corresponding tumour immune microenvironment of these three cases, because it’s a small cluster of three cases, and we found some similarities in terms of CD3 positive tumour infiltrating lymphocytes and PD-L1 levels.
Then we decided to compare this group with another group with similar immune microenvironment characteristics but opposite survival outcomes in order to understand whether there are specific radiomic features with an impact on overall survival. So in order to identify a signature because this is our aim.
Did you find a signature?
We found a sort of signature that needs to be validated for sure, also in terms of a larger group of patients. Because we found 187 CT derived radiomic features which are differentially expressed by these two groups of patients and then we selected the twenty top up- and the twenty top downregulated according to signal to noise ratio, so a specific statistical test. Then the last step of our study was to define whether these signature-related CTRFs had an impact on the overall population of our patients. Surprisingly, or maybe not so much, we found that, yes, there are two specific radiomic signatures, which are large dependence emphasis and cluster tendency, which have an impact on the overall survival of our entire cohort of patients. So, to some extent, I think it’s a good result.
What can we now do with this knowledge?
We can implement the risk stratification model because if we will be able to add these CT-derived radiomic features, or maybe not all these CT-derived but specific CT parameters, and immune parameters to our risk stratification models based on pTNM or other recognised models, we can reach a new level of real knowledge in terms of prognosis of small cell lung cancer patients.
Is there anything you’d like to add?
In addition to immune parameters we are also studying genomic parameters of these specific cases, just to have an idea because the impact of tumour mutational burden or specific gene expression signature and whatever. So our results are still preliminary but we found something interesting also in terms of genomic expression profiles so we will see.