MRDFF: A deep forest based framework for CT whole heart segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Automatic whole heart segmentation is crucial for cardiovascular disease research and treatment.
• Existing methods may struggle with class imbalance and accuracy in segmenting complex cardiac substructures.

Purpose of the Study:

• To propose an improved Deep Forest framework for accurate and efficient whole heart segmentation.
• To address challenges like class imbalance and enhance the segmentation of cardiac substructures.

Main Methods:

• Introduced a two-stage Multi-Resolution Deep Forest Framework (MRDFF).
• Employed binary classification for initial heart region extraction to mitigate class imbalance.
• Utilized hybrid feature fusion, multi-resolution fusion, and multi-scale fusion techniques.
• Subdivided the extracted heart region in the second stage for detailed substructure segmentation.

Main Results:

• The MRDFF achieved comparable accuracy to existing neural network models on the MM-WHS dataset.
• The proposed model demonstrated significantly reduced training time, approximately half that of neural networks.
• The two-stage approach effectively handled class imbalance and improved substructure segmentation.

Conclusions:

• The Multi-Resolution Deep Forest Framework offers an efficient and accurate solution for automatic whole heart segmentation.
• This approach shows promise for advancing cardiovascular disease research by improving segmentation speed and accuracy.
• The MRDFF framework provides a viable alternative to deep learning models for cardiac image analysis.