DMFF-Net: A dual encoding multiscale feature fusion network for ovarian tumor segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Ovarian cancer poses a significant threat to women's reproductive health.
• Accurate tumor segmentation is crucial for effective diagnosis and treatment planning.
• Current automatic segmentation methods struggle with capturing fine tumor details.

Purpose of the Study:

• To propose an advanced deep learning model for precise ovarian tumor segmentation.
• To address the limitations of existing methods in segmenting intricate tumor details.
• To improve the accuracy and reliability of automated medical image analysis for ovarian cancer.

Main Methods:

• Development of a dual encoding based multiscale feature fusion network (DMFF-Net).
• Utilized dual encoding paths with residual and dense aggregation blocks for diverse feature extraction.
• Implemented a multiscale feature fusion block to enhance feature information and reduce loss.
• Incorporated coordinate attention in the decoding stage for accurate information capture.

Main Results:

• The DMFF-Net demonstrated superior performance in segmenting ovarian tumor details compared to existing algorithms.
• The proposed method achieved high accuracy in lesion detail segmentation.
• The network also showed strong performance in other medical image segmentation tasks.

Conclusions:

• The DMFF-Net effectively improves the segmentation of ovarian tumor details.
• This advancement offers a promising tool for enhancing ovarian cancer diagnosis.
• The network's versatility suggests potential applications in broader medical imaging segmentation challenges.