Classification of brain lesions using a machine learning approach with cross-sectional ADC value dynamics

Area of Science:

• Neuroimaging
• Radiology
• Machine Learning in Medicine

Background:

• Diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) values are crucial for assessing brain pathologies.
• Current methods for analyzing ring-enhancing lesions (RELs) using ADC values have limitations.
• Standardization of ADC map assessment for RELs is needed.

Purpose of the Study:

• To develop an image analysis method for recognizing RELs irrespective of absolute ADC values or defined regions of interest.
• To evaluate a novel line of interest (LOI) approach for ADC map analysis in differentiating brain abscesses from glioblastomas (GBM).

Main Methods:

• A line of interest (LOI) was marked on ADC maps to encompass all compartments of RELs.
• Machine learning algorithms (k-NN and SVM with Gaussian kernel) were employed to analyze LOI data.
• Diagnostic performance of selected parameters was assessed using these models.

Main Results:

• The k-NN model achieved 80% correct classification for abscesses and 100% for GBM.
• The SVM model yielded similar high accuracy in differentiating between abscesses and GBM.
• The LOI assessment demonstrated significant diagnostic ability.

Conclusions:

• The proposed LOI method offers a new, standardized approach for evaluating ADC maps in various RELs.
• This technique can improve the non-invasive assessment and differentiation of brain pathologies.
• Further exploitation of ADC value potential in neuro-oncology and infectious diseases is suggested.