DCE-MRI pharmacokinetic parameter maps for cervical carcinoma prediction

Area of Science:

• Biomedical Imaging
• Machine Learning in Oncology
• Pharmacokinetics

Background:

• Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) provides pharmacokinetic parameters for interpreting tissue angiogenesis.
• Cervical carcinoma diagnosis often relies on invasive methods; non-invasive imaging biomarkers are needed.

Purpose of the Study:

• To develop and evaluate machine learning approaches for cervical carcinoma prediction using DCE-MRI pharmacokinetic parameters.
• To compare the efficacy of individual parameters versus combined parameters for cancer detection.

Main Methods:

• Pharmacokinetic parameters were estimated from DCE-MRI data.
• Support Vector Machines (SVMs) were used to combine parameters from pharmacokinetic models.
• A novel method, APITL (artificial pharmacokinetic images for transfer learning), was developed to leverage comprehensive pharmacokinetic information.
• A 'winner-takes-all' strategy consolidated slice-wise predictions to subject-wise predictions.

Main Results:

• The extracellular extravascular space volume fraction (Ve) showed high discriminative power for differentiating cancerous from normal cervical tissue.
• The SVM approach improved accuracy by approximately 10% compared to individual parameters.
• The APITL method achieved a subject-wise prediction accuracy of 94.3%, outperforming SVM.

Conclusions:

• APITL demonstrates high accuracy in predicting cervical carcinoma from DCE-MRI data.
• The developed machine learning approaches, particularly APITL, show significant potential for clinical application in non-invasive cervical cancer diagnosis.