Clear and non-clear cell renal cells carcinomas can now be distinguished using a radiomics model based on enhanced computed tomography.
The kidney is roughly the size of a smartphone and makes up no more than 0.5 per cent of our total body weight. Yet, despite their small size, they are responsible for filtering 45 gallons of our blood each day. Given such heavy responsibilities, it comes with no surprise that our kidneys are as vulnerable as they are powerful, and especially at risk of cancers.
Renal cell carcinoma (RCC) is the most common form of malignant tumour afflicting the kidney, with a high metastasis rate of 20 per cent. RCC is classified as either clear cell RCC (ccRCC) or non-clear cell RCC (non-ccRCC). Presently, enhanced computed tomography and renal biopsies are commonly used to diagnose renal cancers. However, misdiagnoses and sampling errors are still prevalent in clinical practice, and conducting biopsies can place patients at risk of metastasis and haemorrhage.
To develop safer and more accurate diagnostics, a team of researchers from the Affiliated Hospital of Hebei University has recently created predictive radiomics models that can determine possible radiogenomics links between the imaging features and a key ccRCC driver gene—the von Hippel-Lindau gene mutation. Using enhanced computed tomography images of renal cell carcinoma, the team conducted 3D analyses of tumour images based on the volume of the interest and extracted different radiomic features to develop radiomics models that distinguish ccRCC from non-ccRCC.
Radiomics involves extracting information from medical images using advanced feature analysis and converting the features into collectable radiomics data to diagnose the nature of lesions. Studies have shown that radiomic features can be highly reliable in distinguishing tumour types and predicting survival in different types of cancer.
In this novel study, scientists have gathered data from 190 RCC cases and randomly divided them into two groups for training and testing sets in a ratio of 7:3 respectively. After collecting 396 radiomic features and analysing the correlation between these features, Univariate Logistics, and Multivariate Logistics, the team ultimately selected four relatively stable features and applied three machine learning models to build their predictive model for distinguishing RCC subtypes. The machine learning models used included Random Forest, Support Vector Machine, and Logistic Regression.
To evaluate the performance of the Random Forest, Support Vector Machine, and Logistic Regression models, the team calculated values of area under the curve, sensitivity, and specificity of the models and compared them against one another and with the diagnosis made by radiologists. Higher values of the area under the curve indicate better performance to differentiate RCC subtypes in the cortical phase of enhancement.
Based on their findings, it was revealed that the radiomics results were better than those achieved by radiologist diagnosis, with Support Vector Machine showing the highest sensitivity of 1.0, and the Logistic Regression exhibiting the greatest specificity of 0.692. Furthermore, the Random Forest model demonstrated the highest area under the curve value of 0.909, reflecting its promising ability to distinguish RCC subtypes.
Besides showing promise to solve clinical problems that usually require challenging and subjective interpretations, these radiomics programmes also bear great potential to surpass traditional enhanced CT imaging data, where imaging radiomics can show specific characteristics to achieve diagnostic results that outmatch traditional medical imaging data. In future, better-designed radiomics trials are expected to take place to validate the use of this predictive model to aid radiological diagnosis of renal cancer and differentiate ccRCC from non-ccRCC. [APBN]
Source: Wang et al. (2021). Radiomics models based on enhanced computed tomography to distinguish clear cell from non-clear cell renal cell carcinomas. Scientific Reports, 11, 13729.