DMGM: deformable-mechanism based cervical cancer staging via MRI multi-sequence

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Oncology

Background:

• Cervical cancer staging relies on accurate interpretation of multi-sequence MRI.
• Simultaneous analysis of multiple MRI sequences presents challenges for clinicians.
• Computer-aided diagnosis systems offer potential for improved information integration.

Purpose of the Study:

• To develop a deep learning model for cervical cancer staging using multi-sequence MRI.
• To address challenges of limited sample sizes and overfitting in medical image analysis.
• To create an auxiliary diagnostic tool for improved cervical cancer staging.

Main Methods:

• Utilized a deformable convolutional layer and a novel deformable ConvLSTM module.
• Implemented a sequence enhancement strategy to diversify samples and mitigate overfitting.
• Developed the deformable multi-sequence guidance model (DMGM) for auxiliary diagnosis.
• Validated the model using multi-modal data from BraTS 2019.

Main Results:

• The deformable ConvLSTM module and DMGM demonstrated effectiveness in cervical cancer staging.
• The model adapted to deformation mechanisms and synchronized asynchronous scan sequences.
• Overfitting issues were addressed, enhancing performance in small dataset scenarios.
• Strong generalization capabilities were validated on an external dataset.

Conclusions:

• The DMGM is the first deep learning model for multi-sequence MRI analysis in cervical cancer staging.
• The model shows significant potential for improving deep learning applications in medical diagnostics.
• Effective staging was achieved even with limited datasets, offering a promising tool for oncologists.