---

Learning Objectives

By completing this assignment, you will:

• Understand the DICOM file format and its importance in medical imaging
• Read and extract metadata from DICOM files using pydicom
• Visualize medical images with proper window/level settings
• Work with DICOM coordinate systems and spatial relationships
• (Advanced) Handle radiation therapy DICOM objects (RT Structure, RT Dose)

---

Background

DICOM (Digital Imaging and Communications in Medicine) is the universal standard for medical imaging. Every CT scan, MRI, X-ray, and most other medical images you’ll encounter in clinical AI are stored as DICOM files.

Understanding DICOM is essential because:

• The metadata is often more valuable than the pixels — patient info, acquisition parameters, spatial coordinates
• Coordinate systems matter — a model trained on one scanner may fail on another if you ignore geometry
• Window/level affects what you see — the same CT data looks completely different for lung vs. bone vs. soft tissue

---

Instructions

Part 1: DICOM Basics (30 points)

1.1 Reading DICOM Files (10 pts)

Load a CT DICOM file and extract key metadata:

• Patient ID, Study Date, Modality
• Image dimensions (Rows, Columns)
• Pixel spacing and Slice thickness
• Print a summary of the DICOM header

1.2 Understanding DICOM Tags (10 pts)

DICOM uses a tag system (Group, Element) to organize data. For the provided CT file:

• Look up 5 DICOM tags using the DICOM Standard Browser
• Explain what each tag represents and why it matters for AI applications
• Document your findings in the code comments

1.3 Pixel Data Access (10 pts)

Extract and examine the pixel array:

• Get the pixel array from the DICOM file
• Print its shape, dtype, min, max values
• Explain the difference between stored values and Hounsfield Units (HU)
• Apply the rescale slope and intercept to convert to HU

---

Part 2: Visualization & Window/Level (30 points)

2.1 Basic Visualization (10 pts)

Display a CT slice using matplotlib:

• Show the image with a grayscale colormap
• Add a colorbar showing HU values
• Add appropriate title and axis labels

2.2 Window/Level for Different Tissues (15 pts)

The same CT data looks completely different depending on window/level settings. Create a figure showing the same slice with four different window/level presets:

Preset	Window Width (W)	Window Level (L)	Use Case
Lung	1500	-600	Lung parenchyma, airways
Soft Tissue	400	40	Abdomen, organs
Bone	2000	400	Skeletal structures
Brain	80	40	Brain parenchyma

Display all four in a 2x2 subplot figure.

2.3 Custom Window/Level Function (5 pts)

Write a reusable function apply_window_level(image, window, level) that:

• Takes a HU image and window/level parameters
• Returns a normalized image (0-255) suitable for display
• Handles clipping appropriately

---

Part 3: DICOM Geometry & Coordinates (25 points)

3.1 Image Position and Orientation (10 pts)

For medical imaging AI, understanding spatial relationships is critical. Extract and explain:

• ImagePositionPatient — Where is the top-left pixel in 3D space?
• ImageOrientationPatient — How is the image oriented?
• What coordinate system does DICOM use (LPS vs RAS)?

3.2 Loading a CT Series (15 pts)

A CT scan consists of many slices. Write code to:

• Load all DICOM files from a directory
• Sort them by slice location (hint: SliceLocation or ImagePositionPatient[2])
• Stack them into a 3D numpy array
• Print the resulting volume dimensions and voxel spacing (dx, dy, dz)

---

Part 4: Advanced Track — Radiation Therapy DICOM (25 points EXTRA CREDIT)

4.1 RT Structure Set Basics (8 pts)

RT Structure Sets (RTSTRUCT) contain contoured regions of interest (ROIs) like organs and tumors.

• Load the provided RT Structure file
• List all structure names and their ROI numbers
• Extract the contour data for one structure (e.g., “PTV” or “Bladder”)
• Explain the format of contour data (how are the points stored?)

4.2 RT Dose Visualization (8 pts)

RT Dose files contain 3D dose distributions from radiation treatment planning.

• Load the RT Dose file
• Extract the dose grid and its spatial information
• Display a dose colorwash overlaid on a CT slice
• Use an appropriate colormap (e.g., jet or hot) with transparency

4.3 DVH Calculation (9 pts)

A Dose-Volume Histogram (DVH) is a fundamental tool in radiation oncology, showing what percentage of a structure receives at least a given dose.

• Using the RT Structure contours and RT Dose grid
• Calculate the DVH for one structure
• Plot the cumulative DVH (Volume % vs Dose in Gy)
• Report D95 (dose covering 95% of the volume) and V20 (volume receiving ≥20 Gy)

---

Submission via GitHub

1. Complete your work in hw3_dicom.py
2. Commit your changes with meaningful messages:

   git add hw3_dicom.py
   git commit -m "Complete Part 2: window/level visualization"
   

3. Push to GitHub before the deadline:

   git push
   

Commit Expectations

• Multiple commits showing your progress
• Meaningful commit messages describing what you did
• Working code in your final commit

---

Grading Rubric

Core Track (Required)

Component	Points
Part 1: DICOM Basics	30
1.1 Reading DICOM files	10
1.2 Understanding DICOM tags	10
1.3 Pixel data access	10
Part 2: Visualization	30
2.1 Basic visualization	10
2.2 Window/level presets	15
2.3 Custom W/L function	5
Part 3: Geometry	25
3.1 Position and orientation	10
3.2 Loading CT series	15
Git Workflow	15
Multiple meaningful commits	8
Clear commit messages	7
Core Total	100

Advanced Track (Extra Credit)

Component	Points
4.1 RT Structure basics	8
4.2 RT Dose visualization	8
4.3 DVH calculation	9
Advanced Total	+25

---

Resources

• DICOM Standard Browser — Look up any DICOM tag
• pydicom Documentation
• Radiopaedia: Windowing (CT) — Window/level explanation
• DICOM Coordinate System — LPS coordinate system

---

Tips

• Start with Part 1 — Make sure you can read files before trying to visualize
• Use the DICOM browser — When you see an unfamiliar tag, look it up!
• Window/level is just contrast adjustment — Don’t overthink it
• The Advanced Track is genuinely hard — Only attempt after completing Parts 1-3
• Commit early and often — Don’t wait until you’re done

---

Sample Data

Your repository includes sample DICOM data in the data/ directory:

• data/ct/ — A small CT series (10-15 slices)
• data/rt/ — RT Structure and RT Dose files (for Advanced Track)

This data is de-identified and safe to use for educational purposes.