Document Type
Event
Start Date
23-4-2023 5:00 PM
Description
Brain tumors are a common type of cancer, and they do not discriminate based on gender, age, or ethnicity. That said, the severity and type of tumor vary among the diagnosed individual, cancerous or benign. The three most common types for diagnosed individuals are glioma, meningioma, and pituitary. Even so, identifying the type of tumor can be an arduous process for both the doctor and the patient, but one technique known as Convolutional Neural Network (CNN) has been particularly effective in expediently and reliably determining the type of brain tumor. A CNN is a type of neural network that can independently extract imaging features (e.g., face shape, eye color of images of faces), one of these models is known as ResNet101, which contains several residual layers. Through these layers, data can be spread forward, and retain information about the findings of the patient's images. The ResNet model layers were frozen in this case for the first 10 epochs, then when the model had 50 of the layers of the base unfrozen which resulted in the training accuracy being 98% after 10 more epochs and the test accuracy being 96% after it finished. The conclusion drawn is that the results mean that they have shown that they have better than most existing models for this application, due to that a majority of models can only achieve the low 90%.
GC-345 BrainNet: Using Deep Learning to Classify Brain Tumors
Brain tumors are a common type of cancer, and they do not discriminate based on gender, age, or ethnicity. That said, the severity and type of tumor vary among the diagnosed individual, cancerous or benign. The three most common types for diagnosed individuals are glioma, meningioma, and pituitary. Even so, identifying the type of tumor can be an arduous process for both the doctor and the patient, but one technique known as Convolutional Neural Network (CNN) has been particularly effective in expediently and reliably determining the type of brain tumor. A CNN is a type of neural network that can independently extract imaging features (e.g., face shape, eye color of images of faces), one of these models is known as ResNet101, which contains several residual layers. Through these layers, data can be spread forward, and retain information about the findings of the patient's images. The ResNet model layers were frozen in this case for the first 10 epochs, then when the model had 50 of the layers of the base unfrozen which resulted in the training accuracy being 98% after 10 more epochs and the test accuracy being 96% after it finished. The conclusion drawn is that the results mean that they have shown that they have better than most existing models for this application, due to that a majority of models can only achieve the low 90%.