Imaging Omics Analysis for Brain Tumors

Brain tumors are tumors that grow in the cranial cavity, and high-grade brain tumors usually exhibit aggressive growth behavior with a high rate of local recurrence. Moreover, there is a statistically significant difference in the prognosis of patients with high-grade and low-grade brain tumors. Therefore, accurate tumor grading is crucial, which affects patient treatment planning and prognosis management. Conventional imaging is not sufficient to accurately distinguish between high-grade and low-grade brain tumors. Diagnostic imaging is essential for the early detection and monitoring of brain tumors and helps researchers to design more appropriate treatment plans.

We provide imaging omics analysis services for brain tumors

A histopathological diagnosis is an invasive approach limited by the specific location of brain tumors and the inability to assess the heterogeneity of the entire tumor with a pathological diagnosis of local tissues. To overcome this drawback, Alfa Cytology offers our clients brain tumor imaging omics analysis services that can extract high-throughput quantitative features from medical images that reflect brain tumor heterogeneity. We convert them into high-dimensional data, which can subsequently be mined to discover their correlation with tumor histological features reflecting underlying genetic mutations and brain tumors.

The general analysis flow

Extraction of high-throughput features is the core step of our imaging omics analysis, and there are four types of commonly extracted features that we use.

• Shape-based features, such as volume, surface area, densities, etc.
• First-order statistics are usually based on histograms, including mean, entropy, skewness, kurtosis, etc.
• Second-order features, including gray level co-occurrence matrix (GLCM), gray-level run-length matrix (GLRLM), etc.
• Higher-order features are filters applied to the image to infer repetitive or non-repetitive patterns. We commonly use filters such as Laplacian of Gaussian (LOG), wavelet transform (WT), and split dimensionally.

Other computational imaging features such as local binary pattern (LBP) and scale-invariant feature transform (SIFT) are also used by us for image characterization.

Predictive model creation methods

The ultimate goal of the imaging omics analysis we provide for our clients is to build a predictive model of clinical outcomes with specific features. The modeling methods we typically use include supervised, semi-supervised and unsupervised learning. Which approach we take will depend on the type of sample you provide and your analysis needs.

• The methods we use in supervised learning include support vector machines (SVM), lasso logistic regression, and random forest.
• The classifiers we use in unsupervised learning include the K-means algorithm, Gaussian mixture clustering, consensus clustering, etc.
• Semi-supervised learning includes an unsupervised feature learning phase and a supervised model training phase.

Alfa Cytology helps researchers achieve preoperative diagnosis and prognosis prediction of brain tumor disease by establishing valuable imaging omics analysis processes and standards, and we can extract a large number of quantitative features from complex clinical imaging arrays. Please contact our staff to submit your analysis requirements.