Improved U-Net based on cross-layer connection for pituitary adenoma MRI image segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuro-oncology

Background:

• Pituitary adenomas are common neuroendocrine tumors.
• Accurate segmentation of pituitary adenomas in MRI is challenging due to image noise and blurred edges.
• Existing methods struggle with precise lesion localization and outlining.

Purpose of the Study:

• To develop a novel deep learning framework for enhanced pituitary adenoma MRI segmentation.
• To address limitations of blurred edges, high noise, and tissue similarity in pituitary adenoma MR images.
• To improve the accuracy and reliability of pituitary adenoma detection and delineation.

Main Methods:

• A U-Net based deep learning framework incorporating a novel cross-layer connection for multi-scale feature capture.
• Implementation of a full-scale skip structure to leverage multi-layer information.
• Utilization of an improved inception-dense block to enhance receptive field effectiveness and network depth.
• Development of a hybrid loss function combining binary cross-entropy and Jaccard losses to manage small and unbalanced datasets.

Main Results:

• The proposed deep learning framework achieved superior performance in pituitary adenoma MRI segmentation compared to existing algorithms.
• Quantitative metrics demonstrated high accuracy: Dice coefficient of 88.87%, Intersection over Union (IoU) of 80.67%, Matthews correlation coefficient (Mcc) of 88.91%, and precision of 97.63%.
• The method proved effective even with a limited dataset of 500 lesion images from 30 patients.

Conclusions:

• The novel deep learning framework significantly improves pituitary adenoma MRI segmentation accuracy.
• The proposed architecture and loss function effectively overcome challenges associated with image quality and data imbalance.
• This advancement holds promise for more precise diagnosis and treatment planning for pituitary adenomas.