Automatic Detection and Segmentation of Mitochondria from SEM Images using Deep Neural Network

Area of Science:

• Neurobiology
• Cell Biology
• Biomedical Imaging

Background:

• Mitochondrial structure and function are critical in neurobiology.
• High-resolution Scanning Electron Microscopy (SEM) enables detailed mitochondrial imaging.
• Accurate detection and segmentation of mitochondria in EM images are challenging due to complex backgrounds.

Purpose of the Study:

• To develop an improved deep neural network for precise mitochondrial detection and segmentation in EM images.
• To enhance the accuracy of automated mitochondrial analysis in neurobiological research.

Main Methods:

• Proposed an improved Mask R-CNN deep neural network architecture.
• Incorporated morphological processing and mitochondrial contextual information for result refinement.
• Validated the method on FIB-SEM and ATUMSEM datasets.

Main Results:

• The proposed deep neural network achieved comparable performance to state-of-the-art methods.
• Demonstrated effective detection and segmentation of mitochondria in complex EM images.
• Showcased the utility of contextual information for improving segmentation accuracy.

Conclusions:

• The developed deep neural network offers an effective solution for mitochondrial detection and segmentation.
• This method advances the analysis of mitochondrial structure in neurobiology.
• The approach shows promise for improving high-resolution biological image analysis.