Normal Image vs DICOM Image
For example, in MRI, data is acquired in the frequency domain (k-space) and then reconstructed into a spatial image (DICOM) using the Inverse Fourier Transform.
1. Normal Image (JPG, PNG, etc.)
Examples: .jpg, .png, .jpeg
What it contains:
- Only pixel values (visual image)
- Designed for display
Key characteristics:
- Usually RGB (3 channels)
- Pixel values: typically 0–255
- No medical meaning attached
- Lightweight and easy to use
Example shape:
[H, W, 3] → RGB image
2. DICOM Image (Medical Imaging)
Example: .dcm
DICOM = Digital Imaging and Communications in Medicine
What it contains:
- Image (scan)
- Patient info (name, age, ID)
- Scan parameters (scanner type, resolution, modality)
- Metadata (very important)
Key Differences
| Feature | Normal Image | DICOM Image |
|---|---|---|
| Purpose | Display | Medical diagnosis |
| Format | JPG, PNG | .dcm |
| Channels | RGB (3-channel) | Usually grayscale (1-channel) |
| Pixel values | 0–255 | Real physical values (e.g. intensity, SUV) |
| Metadata | No | Yes (extensive) |
| File size | Small | Larger |
| Usage | Web, apps | Hospitals, research |
Important Concept (Interview Focus)
In DICOM:
- Pixel values are not just colors
- They represent real physical quantities
Examples:
- CT → Hounsfield Units (HU)
- PET → metabolic activity (SUV)
img = dcm.pixel_array
This is not just an image — it is medical data.
Why You Can’t Treat DICOM Like JPG
cv2.imread("image.jpg") # normal image
pydicom.dcmread("scan.dcm") # DICOM image
- JPG → already visualized
- DICOM → requires processing and normalization
Why Convert to Grayscale
- DICOM images are usually single-channel
- Some may appear multi-channel → need standardization
if img.ndim == 3:
img = img.mean(axis=-1)
Simple Analogy
- Normal image = Photo
- DICOM image = Medical report + scan
Summary
- Normal images are for display
- DICOM images are for analysis and diagnosis
- DICOM contains rich metadata and real-world values
- Used in radiomics, deep learning, and medical AI