Deep Learning using Pre-Brachytherapy MRI to Automatically Predict Applicator Induced Complex Uterine Deformation.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Brachytherapy (BT) for cervical cancer involves applicator insertion, which deforms the uterus.
• Accurate prediction of uterine deformation is crucial for effective radiation dose delivery.
• Current methods may lack precision in predicting applicator-induced anatomical changes.

Purpose of the Study:

• To develop and validate a deep learning (DL) algorithm for predicting uterine shape and location changes during brachytherapy.
• To automate the segmentation of the uterus in pre-treatment MRI scans.
• To quantify uterine deformation caused by intrauterine/intravaginal applicators.

Main Methods:

• Utilized paired pelvic MRI datasets (pre-BT and at-BT) from 92 cervical cancer patients.
• Developed a DL algorithm using a CNN (Inception V4) with autoencoders for initial uterine segmentation.
• Employed a modified U-net model with transfer learning to predict at-BT uterine shape from pre-BT MRI.
• Quantified deformations using the free-form deformation method.

Main Results:

• The DL algorithm achieved an average Dice Coefficient (DC) of 94.1% and Hausdorff Distance (HD) of 4.0 mm for pre-BT uterus segmentation.
• The modified U-net model predicted at-BT uterus shape with a DC of 88.1% and HD of 5.8 mm.
• Mean uterine surface displacement was 25.0 mm, indicating significant applicator-induced deformation.

Conclusions:

• The novel DL algorithm accurately predicts applicator-induced uterine deformation from pre-treatment MRI.
• This automated prediction tool can aid in personalized brachytherapy treatment planning.
• The framework promises improved clinical decision-making and dosimetric outcomes in cervical cancer treatment.