Construction of VGG16 Convolution Neural Network (VGG16_CNN) Classifier with NestNet-Based Segmentation Paradigm for...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Oncology
• Radiology

Background:

• Brain metastases (BMs) are common in metastatic cancer (MC), necessitating accurate diagnosis for treatment planning and radiotherapy.
• Automated BMs (ABMS) diagnosis faces challenges with small lesions and distinguishing true metastases (MtS) from false positives.

Purpose of the Study:

• To enhance brain metastases classification performance by developing a novel VGG16 convolutional neural network (CNN) framework.
• To integrate temporal prior data for improved automated brain metastases detection.
• To improve the diagnosis of minute metastases and calibrate sensitivity and specificity.

Main Methods:

• Utilized the NestNet framework for metastases localization and segmentation.
• Employed the VGG16 convolutional neural network for classification.
• Developed a novel loss function using the weighted softmax function (WSF) for enhanced small MtS diagnosis.

Main Results:

• The VGG16_CNN framework demonstrated high confidence in differentiating positive MtS from candidates.
• Achieved 93.74% accuracy, 92% precision, 92.1% recall, and 67.08% F1-score.
• The VGG16_CNN performance was comparable to advanced methodologies like moU-Net, DSNet, and U-Net.

Conclusions:

• The VGG16_CNN is effective for reinforcing specialist analysis in actual medical practice for brain metastases detection.
• The proposed method offers a reliable tool for differentiating true metastases, reducing the need for extensive specialist review.
• The integration of temporal prior data and WSF enhances the diagnostic capability for small and challenging brain metastases.