MRI-Based Deep Learning Segmentation and Radiomics of Sarcoma in Mice.

Area of Science:

• Oncology
• Medical Imaging
• Radiomics

Background:

• Small-animal imaging offers noninvasive, longitudinal insights into new cancer therapies.
• Variability in image analysis techniques can lead to inconsistent results in preclinical studies.
• Quantitative imaging is crucial for reliable preclinical cancer research.

Purpose of the Study:

• To develop and validate a quantitative imaging analysis pipeline for preclinical cancer therapy studies.
• To assess the utility of radiomics features for predicting tumor recurrence after radiation therapy.
• To reduce bias and improve consistency in analyzing multiparametric MRI data.

Main Methods:

• A genetically engineered mouse model of soft tissue sarcoma was used.
• Magnetic resonance imaging (MRI) was performed before and after radiation therapy (RT).
• An automated pipeline with convolutional neural networks for segmentation and radiomics analysis was applied to multicontrast MRI data.

Main Results:

• The automated pipeline achieved high segmentation accuracy (Dice scores ~0.86).
• Radiation therapy led to increased tumor volumes and heterogeneity one week post-treatment.
• Radiomics features effectively predicted primary tumor recurrence (AUC: 0.79).

Conclusions:

• The developed image processing pipeline enables high-throughput, reduced-bias segmentation of multiparametric MRI data.
• Radiomics analysis of tumor and peritumoral areas can predict recurrence.
• This quantitative approach enhances the understanding of preclinical imaging for novel cancer therapies.