Course Introduction

Medical Imaging Modalities

Figure 1

X-ray image creation schematic.
Schematic of x-ray image creation.

Figure 2

Knee series.image courtesy of Radiopaedia, author and ID on image

Figure 3

Fluorsocopy.image courtesy of Ptrump16, CC BY-SA 4.0 https://creativecommons.org/licenses/by-sa/4.0, via Wikimedia Commons

Figure 4

Graph of sinogram and processed images.
singogram and processed images.

Figure 5

Mitral valve prolapse.
Image of mitral valve prolapse from Cafer Zorkun, MD, PhD on wikidoc.org with creative commons lisence.

Figure 6

K-space.
k-space image.

Figure 7

Graph of k space and processed images.
K space and processed images.

Figure 8

Nuclear Medicine Image.
Nuclear medicine image.

Figure 9

Pathology Image.
Pathology image.

Working with MRI

Figure 1

Nipreps chart

Figure 2

Arrays

Figure 3

MRI slices

Figure 4

T1 weighted

Figure 5

flipped images

Registration and Segmentation with SITK

Figure 1

SITK logo.

Figure 2

SITK Image.
An image in SITK occupies a region in physical space which is defined by its meta-data (origin, size, spacing, and direction cosine matrix). Note that the image’s physical extent starts half a voxel before the origin and ends half a voxel beyond the last voxel.

Figure 3

Isotropic vs non-isotropic images.
The same image displayed with a viewer that is not aware of spatial meta-data (left image) and one that is aware (right image). The image’s pixel spacing is (0.97656, 2.0)mm.

Figure 4

Spatial relationship in images.
Two images with exactly the same pixel data, positioned in the world coordinate system. In SITK these are not considered the same image, because they occupy different spatial locations.

Figure 5

Slice and cmaps example.

Figure 6

Multiple slices example.

Figure 7

Operations examples.

Figure 8

Slice with grid mask.

Figure 9

Digital x-ray image.

Figure 10

Grayscale x-ray image.

Figure 11

Non-isotropic slices example.

Figure 12

CT and MRI volumes before being aligned.

Figure 13

CT and MRI volumes overimposed.

Figure 14

Metrics across iterations.

Figure 15

CT and MRI volumes aligned.

Figure 16

T1 MRI scan, Z slices.

Figure 17

Basic thresholding methods.

Figure 18

Brain lateral ventricle.

Figure 19

Initial seed.

Figure 20

Region growing segmentations.

Figure 21

Confidence connected after morphological closing.

Figure 22

Level-set segmentation.

Figure 23

Y-axis segmentation.

Preparing Images for Machine Learning

Figure 1

CXR examples

Figure 2

augmented chest x-ray different sizes

Figure 3

augmented chest x-ray

Figure 4

augmented by shear chest x-ray

Figure 5

augmented by waves chest x-ray

Figure 6

T1 v T3

Working with Pathology Images

Figure 1

Gross and histopathology

Figure 2

Thumbnail histopathology

Figure 3

Pyramidal file structure and tiling are quite challenging to understand, and different file formats can actually implement the concepts a bit differently. Pyramids are actually a more general concept in image processing. The key concept about them to understand are that a pyramidal image simply represents the image at multiple scales, and that could be implemented in many different ways. Usually we don’t create such pyramids by hand, but rather, you guessed it, algorithmically with code. The pictures below illustrate the concepts involved in tiled pyramical images. File structures Pyramidal histopathology

Anonymizing Medical Images

Figure 1

Identifiable ultrasound
Image from flikr website published with a permissive lisence.

Figure 2

Non-Identifiable blurred ultrasound
Image after blurring in one area.

Figure 3

Non-Identifiable masked ultrasound
Image after masking in one area.

Figure 4

Non-Identifiable cropped ultrasound
Image after crop and resize.

Figure 5

Non-Identifiable head
Images of SITK head.

Figure 6

Home-made deface
Our partial soft tissue stripping.

Figure 7

Home-made deface by grow from seed
Our grown from seed soft tissue stripping.

Figure 8

Defacing examples
Image from “A reproducibility evaluation of the effects of MRI defacing on brain segmentation” by Chenyu Gao, Bennett A. Landman, Jerry L. Prince, and Aaron Carass. The preprint is available here.

Figure 9

jewlery artifact
Case courtesy of Ian Bickle, Radiopaedia.org. From the case rID: 61830

Generative AI in Medical Imaging

Figure 1

Misled image
Image generated by Dr. Candace Makeda Moore prompting Adobe Firely.

Figure 2

Misled image of cats
Image generated by Dr. Candace Makeda Moore prompting Adobe Firely.