DEEP-LEARNING SEGMENTATION OF CT AND MR IMAGES: A PRECIOUS DIAGNOSTIC TOOL FOR CRANIOFACIAL LESIONS
Authors:
Shadi Daoud1,2, Idan Redenski1,2, Adeeb Zoabi1,2, Fares Kablan1,2, Samer Srouji1,2
Affiliation:
1 Galilee College of Dental Sciences, Oral and Maxillofacial Surgery Department, Galilee Medical Center, Nahariya, Israel
2 The Azrieli Faculty of Medicine, Bar-Ilan University, Safed, Israel
Doi: 10.54936/haoms242p35
ABSTRACT:
Introduction: RMedical segmentation entails extracting specific regions of interest from three-dimensional image data, such as Magnetic Resonance Imaging (MRI) or Computed Tomography (CT) scans. The primary objective of segmenting this data is to identify anatomical areas that are necessary for a particular study or analysis, such as lesions. The segmentation process encapsulates a wealth of information regarding the properties of the lesion, and harnessing this information effectively can greatly aid in achieving an accurate diagnosis.
Materials and Methods: We have developed a deep learning algorithm that has been introduced to widely accepted maxillofacial and radiographic references. This algorithm has been specifically designed to integrate lesions’ information and function as a valuable tool for both differential diagnosis and treatment planning.
The algorithm focuses on lesion properties which includes, Thresholding - Hounsfield unit - Lesion surface/borders and Intralesional properties, Morphological characteristics, Single or Multifocal/Size & Volume, Lesion location, Evaluating the effect on surrounding structure. Moreover, Lesion superimposition – used for lesion follow-up.
Our deep learning algorithm underwent thorough examination and preparation by being trained on a dataset comprising 250 intraosseous lesions (CT) and 150 soft tissue lesions (MRI).
Results: Through the utilization of deep learning, our algorithm has achieved an 87% accuracy in differential diagnosis for maxillofacial lesions. While the accuracy of diagnosis for soft tissue lesion on MRI was 82% the accuracy of CT lesion segmentation-based diagnosis was 90%.
Conclusions: Deep learning algorithm is a decision support tool for the radiological assessment images of craniofacial lesions. This emerging tool allows us to expedite the interpretation process and enhancing workflow efficiency leading to improved clinical outcomes. Furthermore, it enhances accuracy and minimizes inconsistencies among different readers.
KEY WORDS:
Authors:
Shadi Daoud1,2, Idan Redenski1,2, Adeeb Zoabi1,2, Fares Kablan1,2, Samer Srouji1,2
Affiliation:
1 Galilee College of Dental Sciences, Oral and Maxillofacial Surgery Department, Galilee Medical Center, Nahariya, Israel
2 The Azrieli Faculty of Medicine, Bar-Ilan University, Safed, Israel
Doi: 10.54936/haoms242p35
ABSTRACT:
Introduction: RMedical segmentation entails extracting specific regions of interest from three-dimensional image data, such as Magnetic Resonance Imaging (MRI) or Computed Tomography (CT) scans. The primary objective of segmenting this data is to identify anatomical areas that are necessary for a particular study or analysis, such as lesions. The segmentation process encapsulates a wealth of information regarding the properties of the lesion, and harnessing this information effectively can greatly aid in achieving an accurate diagnosis.
Materials and Methods: We have developed a deep learning algorithm that has been introduced to widely accepted maxillofacial and radiographic references. This algorithm has been specifically designed to integrate lesions’ information and function as a valuable tool for both differential diagnosis and treatment planning.
The algorithm focuses on lesion properties which includes, Thresholding - Hounsfield unit - Lesion surface/borders and Intralesional properties, Morphological characteristics, Single or Multifocal/Size & Volume, Lesion location, Evaluating the effect on surrounding structure. Moreover, Lesion superimposition – used for lesion follow-up.
Our deep learning algorithm underwent thorough examination and preparation by being trained on a dataset comprising 250 intraosseous lesions (CT) and 150 soft tissue lesions (MRI).
Results: Through the utilization of deep learning, our algorithm has achieved an 87% accuracy in differential diagnosis for maxillofacial lesions. While the accuracy of diagnosis for soft tissue lesion on MRI was 82% the accuracy of CT lesion segmentation-based diagnosis was 90%.
Conclusions: Deep learning algorithm is a decision support tool for the radiological assessment images of craniofacial lesions. This emerging tool allows us to expedite the interpretation process and enhancing workflow efficiency leading to improved clinical outcomes. Furthermore, it enhances accuracy and minimizes inconsistencies among different readers.
KEY WORDS: