PRT-Net: a progressive refinement transformer for dose prediction to guide ovarian transposition

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Cervical cancer patients undergoing pelvic radiotherapy often require ovarian transposition to preserve ovarian function.
• Accurate localization of transposed ovaries is crucial to minimize radiation dose and prevent exceeding limits.
• Determining a safe ovarian position before surgery is essential for effective fertility preservation.

Purpose of the Study:

• To develop and validate a novel AI model for predicting radiation dose distribution in transposed ovaries.
• To assist surgeons in identifying optimal, low-dose regions for ovarian transposition prior to radiotherapy.
• To enhance the preservation of ovarian function in young cervical cancer patients receiving pelvic radiotherapy.

Main Methods:

• Proposed a progressive refinement transformer network (PRT-Net) for multi-scale dose prediction from preoperative CT scans.
• Implemented a deep supervision strategy for refining dose predictions from low to high resolution.
• Utilized a multi-loss function fusion algorithm to optimize prediction accuracy across different loss dimensions.
• Validated the clinical feasibility of the PRT-Net model using actual patient cases.

Main Results:

• PRT-Net successfully generated accurate radiation dose predictions for transposed ovaries.
• The model enabled surgeons to quickly identify low-dose regions for safe ovarian placement.
• Clinical verification confirmed the feasibility and utility of the PRT-Net approach in real-world scenarios.

Conclusions:

• PRT-Net provides a valuable tool for preoperative planning of ovarian transposition in cervical cancer patients.
• The AI-driven dose prediction assists in preventing ovarian dose limits from being exceeded during radiotherapy.
• This method supports improved ovarian function preservation and fertility outcomes for young cancer survivors.