Comprehensive Guide to Image Segmentation, Neural Networks, and 3D Printing in Medical Imaging

What is manual segmentation and its challenge?

it is tedious and requires segmenters to be very precise.

Define interpolation in the context of image segmentation.

the process where the software takes contours drawn manually on slices and fills in the missing contours on the slices between.

How does interpolation improve segmentation efficiency?

It reduces manual work and provides more consistent results by allowing the program to generate intermediate shapes instead of outlining every slice manually.

What is the difference between 2D and 3D interpolation?

2D interpolation computes shapes between slices, while 3D interpolation fills in a continuous surface/volume across all spatial directions.

What is extrapolation in segmentation?

when a computer predicts or extends beyond the range of data provided, used to segment structures like arteries based on trends from existing contours.

What is the risk associated with extrapolation?

it involves guesswork, making it riskier as it may inaccurately predict structures beyond the provided data.

Explain bilinear interpolation.

it involves taking the last pixel value minus the first, dividing by the separation, and adding the resulting n value to the start point to find intermediate values.

What is nearest neighbor interpolation?

type of interpolation that assigns each unknown voxel or pixel the label of the closest known labeled voxel, creating a stepwise blocky segmentation.

How does the smooth filter work for images?

operation that uses a mask to find the mode, median, or mean of pixels in a defined area, reducing noise and smoothing the image.

What is k-means clustering?

an unsupervised machine learning algorithm that organizes data into a specified number of clusters based on proximity to centroids.

Describe the process of k-means clustering.

It begins by selecting random initial centroids, assigning data points to the nearest centroid, recalculating centroids, and repeating until clusters stabilize.

What are common applications of k-means clustering in imaging?

It is used for classifying image pixels, separating soft tissue from bone in CT scans, and clustering tissues in histology or microscopy images.

What is a deep neural network?

a machine learning model made up of many connected layers of artificial neurons that learn to recognize complex patterns from data.

Why are multiple layers important in a deep neural network?

Multiple layers allow the network to build up from recognizing simple patterns to identifying complex concepts, enabling it to learn intricate features.

What is the input for a deep neural network processing images?

can be a 2D image, a stack of slices, or a full 3D volume represented as an array of numbers based on pixel intensities.

What types of patterns do lower layers of a deep neural network detect?

detect simple patterns such as edges, lines, colors, and brightness differences.

What do middle layers of a deep neural network combine?

combine simple patterns into more meaningful features like corners, textures, and small shapes.

What complex concepts do higher layers of a deep neural network recognize?

recognize complex concepts such as organs, tumors, anatomical structures, segmentation masks, and classifications.

What is the function of a convolution layer in a deep neural network?

It applies small filters or kernels that slide across the image to detect patterns such as edges, textures, and shapes.

What does a deconvolution layer do in a CNN?

It unsamples feature maps to increase spatial resolution, recovering details lost during pooling and combining high-level features with fine spatial information.

What is the purpose of a pooling layer in a CNN?

It downsamples feature maps to reduce their size, making the network smaller, faster, and more robust to noise, while focusing on the most important features.

What is the sequence of operations in a CNN?

Convolution -> Pooling -> Unpooling -> Deconvolution

How do CNNs learn features from input data?

They learn their own features based on the input data, eliminating the need for predefined features.

In what scenarios are neural network-based segmentation particularly useful?

When anatomy is complex, boundaries are unclear, or intensities overlap, such as in bone vs. soft tissue separation or tumor segmentation.

What is the definition of Artificial Intelligence (AI)?

a broad category involving techniques that enable computers to emulate human behavior, learning, decision-making, and problem-solving.

How does Machine Learning (ML) relate to AI?

a subset of AI that uses advanced algorithms to detect patterns in large datasets, allowing machines to learn and adapt.

What distinguishes Deep Learning (DL) from Machine Learning?

a subset of ML that uses neural networks with multiple layers to extract high-level features from raw data.

What is Generative AI?

a subset of deep learning models that create content like text, images, or code based on input, using patterns learned from vast datasets.

What is a single neuron in a neural network model? and what does it do

A perceptron that takes multiple input features, multiplies them by a weighing factor, adds a bias term, then computes the weighted sum and applies a nonlinear function to compute its output. The calculated nonlinear output is then passed to the next layer

What is the role of unpooling in a neural network?

process that reverses the pooling effect, increasing resolution without adding new features or complexity.

What is deconvolution in the context of neural networks?

process that is a learnable unsampling process that applies filters to increase spatial resolution and reconstruct features.

What are the sequential steps to go from a DICOM file to a printed file in an FDM printer?

Preprocessing, segmentation, post-processing, ensuring accurate anatomy, STL generation, CAD processing, slicing, and exporting the STL file.

What is the function of photopolymerization in SLA printing?

It involves curing liquid resin into solid parts using light, typically involving materials like PEGDA and crosslinkers.

What is the significance of accurate anatomy in 3D printing?

Accurate anatomy ensures that the printed model correctly represents the intended structure, which is crucial for applications like medical modeling.

What is the purpose of segmentation in the context of neural networks?

Segmentation involves separating different features or regions in an image to facilitate analysis or processing.

What is the difference between min pooling and other pooling methods?

Min pooling selects the minimum value from a feature map segment, while other methods like max pooling select the values based on other parameters.

What initiates photopolymerization in PEGDA?

A photoinitiator absorbs UV light and generates free radicals.

What is the role of free radicals in photopolymerization?

They attack carbon-carbon double bonds in acrylate groups, initiating a chain-growth reaction.

What is formed as PEGDA molecules link together during photopolymerization?

A crosslinked hydrogel network.

What is the broad definition of photopolymerization?

A process where light energy initiates a chemical reaction that turns liquid monomers into a solid polymer network.

What are the three main types of 3D printing technologies discussed?

Fused Deposition Modeling (FDM), Stereolithography (SLA), and Selective Laser Sintering (SLS).

How does FDM 3D printing work?

It melts and hardens filament material layer by layer.

What is the main focus of SLA 3D printing?

It uses chemical reactions to harden materials, resulting in less anisotropic features than FDM.

How does SLS 3D printing differ from FDM and SLA?

It sinters material with a laser without generating supports and does not melt unnecessary material.

Why were food coloring dyes used in the study?

They are biocompatible and prevent cytotoxic effects of typical photoabsorbers.

What does the figure showing deformation of printed holes illustrate?

The deformation based on the printed axis and various layer thicknesses.

What does the absorbance spectra of tartrazine indicate?

It encompasses the light source used to initiate crosslinking via LAP photoinitiator.

What was a significant drawback of the study by Grigoryan et al.?

It required a custom printer and used animal subjects that were killed for experimentation.

What physiological tests were performed with the alveolar model?

Tidal ventilation testing, gas exchange performance, and bidirectional blood flow testing.

What materials are commonly used in SLS 3D printing?

Nylon-based materials, polypropylene, and polyurethane.

What material was used in the article by Raza et al.?

PA12 powder with a particle size of 50 microns.

What variables were tested in Raza et al.'s study?

Laser power, scan speed, hatch spacing, and scan length.

What constants were maintained in Raza et al.'s experiments?

Powder bed temperature, ambient temperature, cooling time, layer thickness, and laser spot diameter.

What is the design of experiments (DOE)?

A method to test selected parameters at multiple levels, organized in a grid to explore all possibilities.

what are some disadvantages of using FDM

Anisotropic properties, requires post processing, can require supports

when do you need support structures in FDM

when there are overhangs of less than 45 degrees or hallow structures, or when there is bridging over 5 mm

what is shell and infill

the thickness of the outer perimeter of the printed object

the density and shape of the material that fills the empty spaces in a printed object

what are some infill geometries used in printing

cubic

honeycomb

grid

gyroid

what are 3 of the top materials used in 3d printing (FDM)

PLA

ABS

TPU

what are 3 kinds of resin used in 3d printing (SLS)

standard resin

elastic

tough

biomed clear

What are 3 kinds of material used in SLS

Nylon 12

alumide

PA-GF (glass filled nylon)

What are some advantages and disadvantages of FDM, SLS, and SLA

FDM:

- pros: inexpensive, accessible to most

- cons: anisotropic, needs supports and post processing

SLS:

- pros: easily reuseable material, no supports needed, strong

- cons: high material cost, limited material variety, shrinkage and warping

SLA:

- pros: chemical bonds so parts are isotropic, can make see through parts, high precision

- cons: expensive, extensive post processing, brittle parts

what are some post processing steps for FDM

Sanding, trimming supports, polishing, cold welding, gap filling, coating, painting

What is SLA? How does it work?

stereo lithography. Additive process in which a liquid photopolymer is selectively cured by light activated polymerization. either by a laser or projector depending on the type of printer used (DLP)

What are the different orientations possible with SLA

large cross sectional area, laying flat across print bed or small cross sectional area in which the print is angled down from the print bed

Describe post processing steps for SLA

removal of part from print bed, isopropyl alcohol to remove part from bed, isopropyl bath for both part and print platform, further curing of part, then support removal

additional steps could be sanding and wet sanding, mineral oil finish, spray painting or polishing

What is SLS? How does it work

Selective laser sintering. A laser sinters powder in the cross section shape of the part that is being printed which adheres the layers together

what are some post processing steps to sls

loose powder removal, surface finish aesthetics like dyeing and painting/ lacquering, metal plating

What is material jetting and how does it work?

additive manufacturing process in which droplets of material are selectively deposited and cured on a build plate.

What is binder jetting and how does it work

Additive manufacturing process in which a liquid bonding agent selectively binds regions of a powder bed.

what is thresholding

primary method for segmentation, selecting a range of grey values to isolate or highlight from an overall image based on the contrast of the grey values around it

what is region growing

a user defined seed point that either adds neighboring pixels of similar intensity or does not add a pixel based on dissimilar intensity and the region grow stops

what are the two settings for region growing

6 connectivity - all voxels that share a face are considered

26 connectivity - all voxels that share a face and edge are considered

what is dynamic region growing

combines features of both thresholding and region growing it is based on the connectivity of gv values based on connectivity of the seed point. it can be used to initiate mass segmentation

what is draw profiles line

another way to threshold, displays how the gray value changes along a line within an image, used when you do not know the GV of your part you are trying to segment, normal thresholding is used after figuring out the gv window

what is 3d live wire

a tool in which you choose the automatic contour and highlight the boundaries in saggital and coronal sections until you see a grid pattern

what does gradient magnitude and attraction do in livewire

0% magnitude - contour will be attracted to darker regions, 100% - the contour will be attracted to brighter regions

-3 attraction, all the small inclusions will be included in the mask, 3 attraction, all the inclusions will be excluded from the automatic contour

what is segmentation by local threshold

a tool that helps you to improve already existing segmentation by thresholding zone directly attached to already existing mask, good for thresholding to separate extra anatomy

what is smart brush

tool that allows you to edit an existing mask or create a new one by marking slices on the 2D images with a pixel intensity sensitive brush. Shows a preview of voxels that will be added or removed by hovering the cursor over the mask

what is crop mask

crop tool that works by selecting the region of interest and restricting your segmentation to it

what is split mask

split one mask into multiple smaller masks, potentially uses a combination of pixel intensities and edge detection algorithms to excecute the splitting

what is edit mask

tool that lets you make small, accurate modifications to the mask on a single slice. You can either add or remove a mask on a single slice, you can also use the threshold operation while using this tool. Great for small changes

what is multiple slice edit

tool that lets you edit masks with minimal input, takes advantage of interpolation and accurately captures the curves and shapes based on pixel intensity

what is 3d interpolate

works the same way as multiple slice edit except you have to select regions in two planes to create a 3d structure

what is segment thin bone

tool that lets you close holes that appear eithin cortical bone, line drawn to select holes, holes patched in 3d and pixels shows as red in 2d images

what is the boolean operation minus

a resulting mask is equal to the difference between sets of pixels of masks a and b. the threshold value of the resulting mask is equal to the threshold value of mask A

what is boolean operation unite

a resulting new mask includes the pixels of both the masks a and b. the lower threshold value of the resulting mask is equal to the lowest threshold value for masks A and B. the upper threshold value of the resulting mask is equal to the highest threshold value for masks a and b.

what is boolean operation intersection

a resulting new mask is equal to the common set of pixels shared by both masks a and b (a mutual part of the sets). the lower threshold value of the resulting mask is equal to the higher value of the lower threshold values for masks a and b. the upper threshold value of the mask is equal to the lower one of the threshold values for masks a and b

what is morphology tool erode

the pixels are removed from the edges of an image

what is morphology operation dilate

pixels are added to the edges of an image

what is morphology operation open

operation first performs and erosion immediately followed by a dilation, small edges will be removed or opened and internal holes may become larger

what is morphology operation close

first performs a dilation immediately followed by an erosion. the small internal cavities and gaps on edges will be closed

what is cavity fill

allows to fill the internal gaps of the selected masks and add the filled region to the mask

either automatic with holes selected one by one or automatic where it auto fills all holes across many slices

what is smart fill

tool that auto fills holes by global filling and local filling. define hole closing distance parameter, user defined voxels for filling holes and closing distances

what is smart expand

the initial mask that will be expanded. it will expand until it finds an edge in the image according to the gray value gradients

what is smooth mask

smooths the mask by integrating pixels, opposite of interpolation or performs other methods such as median, mode, or mean