Automatic classification of medical image modality and anatomical location using convolutional neural network

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Radiologic images often use the DICOM standard, which contains vital metadata.
• Converting DICOM images to other formats can result in loss of critical image details and information.
• Efficient handling and storage of large medical image datasets are crucial for both clinical and research applications.

Purpose of the Study:

• To develop and evaluate a Convolutional Neural Network (CNN) architecture for classifying medical images converted to non-DICOM formats.
• To assess the CNN's ability to differentiate between various imaging modalities (CT, MRI) and anatomical locations (abdomen, brain, spine).

Main Methods:

• Four classes of medical images were created: CT abdomen, CT brain, MRI brain, and MRI spine.
• Images were converted from DICOM to JPEG format.
• A proposed CNN architecture was utilized to automatically classify the converted images.

Main Results:

• The CNN achieved excellent overall classification accuracy exceeding 99.5% on both validation and test sets.
• Specificity and F1 scores were above 99% for each image category.
• The model successfully classified images based on imaging modality and anatomical location.

Conclusions:

• CNNs are a promising methodology for classifying non-DICOM medical images.
• This approach can potentially save image processing time and storage space.
• The CNN model demonstrates high accuracy in distinguishing between different medical image types and locations.