A Fully Convolutional Deep Neural Network for Lung Tumor Boundary Tracking in MRI

Area of Science:

• Medical Imaging
• Radiotherapy
• Artificial Intelligence

Background:

• Accurate lung tumor segmentation in MRI is crucial for effective radiation therapy planning.
• Image similarities and respiratory motion complicate manual tumor delineation, making it time-consuming and costly.
• Automated segmentation methods are needed to improve efficiency and accuracy in radiotherapy.

Purpose of the Study:

• To develop and evaluate an automated method for tracking lung tumor boundaries using convolutional neural networks (CNNs).
• To assess the performance of the proposed CNN architecture with a modified Dice metric as the cost function.
• To compare the automated method against expert manual delineations and state-of-the-art approaches.

Main Methods:

• A convolutional neural network (CNN) architecture was employed for automatic tumor boundary tracking.
• A modified Dice metric was utilized as the cost function within the CNN.
• The proposed method was validated on a dataset of 600 MRI images.

Main Results:

• The automated method achieved a high average Dice score of 0.91 ± 0.03.
• The Hausdorff distance was measured at 2.88 ± 0.86 mm, indicating precise boundary delineation.
• The proposed approach demonstrated superior accuracy compared to existing state-of-the-art methods for mobile tumors.

Conclusions:

• The developed CNN-based approach offers an accurate and efficient solution for automatic lung tumor segmentation in MRI.
• This automated method can significantly aid in radiation therapy planning by providing reliable tumor boundary tracking.
• The findings suggest a promising advancement in medical image analysis for cancer treatment.