On this page you will find supplemental data and background information on two student intern posters. For background information regarding 3D printing for the visualization of cell biology at the nanoscale, keep scrolling down on this page. 

Pseudo labels yield better mitochondrial segmentations from FIB-SEM datasets 

Patrick Aquino, Ryan Conrad, Kedar Narayan



Glossary (in terms of my work):

  • Hardening - a forced conversion of probability values (0..1) to binary (0 or 1) based on a cut-off (threshold)
  • Intersection over Union - an evaluation metric used to measure the accuracy of an object detector (see Figure 2, below)
  • Ground Truth - an accurately labeled map (often done manually) against which predictions and results are compared  
  • Neural Network -   a set of algorithms, modeled after the human brain, that are designed to recognize patterns. In this case, we are writing an algorithm that is able to detect mitochondria in an electron microscopic image of cells
  • (Binary) Label Map - a map that is labeled with either 0 (for no) and 1 (for yes), e.g., 0 for each pixel that does not represent a mitochondria and 1 for each pixel that represents a mitochondria
  • Precision - how close the measured values are to each other (See Formula, below)
  • Pseudo Labeling - fine tuning the neural network with the hardened post processed predictions
  • Recall/Sensitivity - measures the proportion of actual positives that are correctly identified as such (See Formula, below)
  • Region of Interest (ROI) - are samples or sub-areas within a data set identified for a particular purpose
  • Specificity - measures the proportion of actual negatives that are correctly identified as such (See Formula, below)
  • Threshold - A cut-off mark that sets all values below threshold (x) as 0s and all values above threshold (x) as 1s (see Figure 1, below)
  • Voxel - a pixel in 3-D 



Note: "Positive" = presence of mitochondria  "Negative" = absence of mitochondria 


    Figure1       Figure 2




3DSlicer - Used for segmentation, creation of ground truth, and prediction clean up

Download Link:

Jupyter Notebook - Used for 2D image printing and statistics computation

Installation Guide:

Python Download:


3-D Printing as a tool to visualize cell biology at the nanoscale 

Hanbin Lee, Kedar Narayan 


Software Used:

3DSlicer: 3DSlicer is the open source software for the image analysis and scientific visualization. It was used for segmentation in this project. 

Download :

CURA: CURA is used to convert .stl files to .gcode files. While .stl files are widely used for 3-D modeling, 3-D printers can’t read .stl files, so we have to convert .stl file to .gcode, which is a programming language used in manufacturing processes. 

CURA is a software specific for Ultimaker 3-D printers. Various 3-D printer companies provide different software that interfaces with their hardware.  

Download :

Ultimaker2: A FDM type 3-D printer used for this project. Any FDM type printers that have build volumes equal to or larger than 22.3 cm × 22.3 cm × 20.5 cm can be used. 

Where to buy:


How to print :

(On the Computer)

1. Make the design using any 3-D modeling software(3DSlicer, Blender, CAD etc)

2. Export design as an .stl file

3. Open .stl file in CURA(for Ultimaker)

4. Save it to SD card (if needed) as a .gcode file


(On the printer)

1. Insert SD card(if relevant)

2. Calibrate the build-plate(Ultimaker2)

    - Spin circular cursor to “Maintenance” on the screen and select 

    - Spin circular cursor to “Build-Plate” on the screen and select 

    - follow instructions listed on the screen

    - use a slide (they are usually 1mm thick) for the 1mm setting

    - adjust the height by twisting the screw under the build-plate 

3. Select print

4. After the printing, cut off the all the support


UltiMaker2 G-Code Files