AI in Diagnostic Radiology Handout PDF

AI in Diagnostic Radiology Recap on Image Data I Arrays It is the representation of a group of numbers in Computer ID Arrays: These are linear structures with a single row or column, representing a sequence of elements. It is called vector 2D Arrays (Matr...

AI in Diagnostic Radiology Recap on Image Data I Arrays It is the representation of a group of numbers in Computer ID Arrays: These are linear structures with a single row or column, representing a sequence of elements. It is called vector 2D Arrays (Matrices): These consist of rows and columns, like a grid 3D Arrays: These extend matrices into three dimensions, Length, Width and Depth  Recap on AI and deep Learning S Machine learning is the core of the AI while Deep Learning is more advanced than machine learning and it imitates the function of the brain S In machine learning, parameters are the values the model learns during training to make predictions. E.g.: in linear regression, the parameters are the coefficients that determine the relationship between input features and the output prediction. The goal is to find the best-fitting line or model that captures the underlying patterns in the data. H.B. Regression is used to predict continuous numerical values while classification predicts category Deep learning It is an artificial neural networ through which you can build complicated networks with millions of parameters with complicated architecture  Convolutional neural network A Convolutional Neural Network (CNN) is a specialized type of neural network primarily used for processing image data. It works by applying multiple layers of filters (convolutions) to the image to capture different features, such as edges and textures, as the data passes through successive layers. CNNs also employ pooling layers to reduce dimensionality and fully connected layers to make final predictions. These networks are highly effective in tasks like image classification and object detection due to their ability to learn spatial hierarchies in data. Padding refers to adding extra pixels (typically zeros) around the input image before applying convolution operations. This is done to maintain the spatial dimensions of the image after convolution or to control the output size.  3D Convolutional neural network In addition to width and height, the third dimension allows 3D CNNs to capture temporal or volumetric information, making them suitable for analyzing sequential data, like volumetric data such as CT or MRI scans. 3D CNNs are effective for tasks like action recognition, medical imaging, and spatial- temporal analysis. The added dimension allows the model to learn more complex patterns from spatial and temporal data.  Data augmentation and preprocessing ❖ Data preprocessing : it's a preparation of raw data for input into a machine learning model. Steps of preprocessing: ❖ Data augmentation: it's a technique used in machine learning to increae the size and diversity of a dataset by modifying already existing data. ❖ It's useful in training models when original data is limited. ❖ Image Data Augmentation: Flipping (horizontal/vertical) Rotation Cropping Adding noise Color adjustments (brightness, contrast, saturation) Scaling or resizing.  Segmentation in radiology ❖ It's the process of partitioning medical images into regions of interest (ROIs) (Highlight certain areas) to isolate specific anatomical structures or pathological areas. ‫بطلب منه يلون أجزاء معينة في األشعة‬ ❖ Help in diagnosis, treatment planning, and image-guided interventions.