Relative location prediction in CT scan images using convolutional neural networks

Area of Science:

• Medical Imaging
• Deep Learning
• Computer Vision

Background:

• Relative location prediction in CT scans is crucial but challenging.
• Traditional machine learning methods lack the required speed and accuracy for medical scenarios.

Purpose of the Study:

• To propose a novel regression model using one-dimensional convolutional neural networks (CNNs).
• To achieve fast and precise relative location prediction in CT scan images.

Main Methods:

• A CNN model utilizing feature vectors from a shape context algorithm as input.
• Pre-processing with z-score normalization and dropout layers to prevent overfitting.
• Model architecture includes convolutional, dropout, and fully-connected layers.

Main Results:

• Achieved excellent performance on a public dataset via 5-fold cross-validation.
• Reported a median absolute error of 1.04 cm and mean absolute error of 1.69 cm.
• Demonstrated rapid prediction time of approximately 2 ms per image.

Conclusions:

• The proposed CNN method enables quick and accurate relative location prediction in CT images.
• Potential to enhance the efficiency of medical picture archiving and communication systems.