AdaptAhead Optimization Algorithm for Learning Deep CNN Applied to MRI Segmentation.

Area of Science:

• Artificial Intelligence
• Machine Learning
• Deep Learning

Background:

• Deep learning, a subset of machine learning, utilizes deep convolution neural networks (DCNNs) for tasks like machine vision.
• Traditional optimization algorithms often struggle with network training accuracy and high computational costs, especially for large datasets.
• DCNNs excel at extracting hierarchical features from large volumes of data.

Purpose of the Study:

• To develop an advanced optimization algorithm, AdaptAhead, for efficient DCNN training.
• To enhance the robustness of DCNN architectures when handling high-volume medical imaging data.
• To improve the accuracy of brain tumor segmentation using deep learning.

Main Methods:

• Developed the AdaptAhead optimization algorithm tailored for DCNNs.
• Validated the algorithm on multi-modality MR images from the BRATS 2015 and BRATS 2016 datasets.
• Compared AdaptAhead's performance against established optimization methods.

Main Results:

• The AdaptAhead algorithm demonstrated improved performance on large datasets compared to other methods.
• Achieved higher accuracy in brain tumor segmentation on the BRATS 2015 and BRATS 2016 datasets, as measured by the Dice similarity metric.
• The algorithm's effectiveness is attributed to its deep and hierarchical feature extraction capabilities.

Conclusions:

• The AdaptAhead optimization algorithm offers a significant advancement for training DCNNs, particularly with large datasets.
• This method provides superior accuracy for medical image segmentation tasks, such as brain tumor identification.
• AdaptAhead represents a robust solution for deep learning applications requiring high-performance optimization.