DICOM vs JPEG: Why File Format Matters in Medical Imaging

The Problem: Screenshots Instead of Data

A referring physician emails you a JPEG screenshot of a patient's knee radiograph. You can see the image. You can form a general impression. But you cannot adjust the window/level, you cannot measure joint space width with calibrated tools, and you cannot feed it to an AI model that requires standardised DICOM input.

This scenario is remarkably common, and it represents a fundamental loss of clinical information. Understanding why requires knowing what each format actually contains — and what it discards.

Bit Depth: 12-Bit Diagnostic Data vs 8-Bit Display

A digital radiograph is typically acquired at 12-bit or 14-bit depth, meaning each pixel stores 4,096 to 16,384 distinct grey levels. This range captures the full spectrum from air (black) through soft tissue to dense cortical bone (white), with enough granularity to distinguish subtle density differences — the faint line of an undisplaced fracture, the slight haziness of early osteophyte formation, the marginal density change of early subchondral sclerosis.

A JPEG file stores 8 bits per channel — 256 grey levels. When a 12-bit DICOM image is exported to JPEG, 94% of the original grey-level information is permanently discarded. Subtle density differences that were clearly separable in the original data merge into identical pixel values and disappear entirely.

This is why DICOM files support windowing: the viewer maps a selected range of the full data (for example, WC:40 WW:400 for soft tissue) onto the display's 256 grey levels. The same pixel data produces entirely different diagnostic images depending on the window settings. A JPEG captures only one window — the data needed for other windows is permanently lost.

Metadata: The Invisible Clinical Record

A JPEG file contains pixels and basic EXIF data (camera model, date, GPS coordinates). A DICOM file contains pixels plus a structured clinical record with hundreds of standardised fields defined by the DICOM standard.

Patient identification (name, ID, date of birth), study information (date, referring physician, accession number), series parameters (slice thickness, pixel spacing, acquisition protocol), and equipment data (manufacturer, model, software version) are all embedded in every DICOM file. This metadata is not optional — it is what connects an image to its clinical context.

The most clinically important metadata field for measurement is Pixel Spacing (DICOM tag 0028,0030). This tag records the physical distance between pixel centres in millimetres. When you measure joint space width in a DICOM viewer, the measurement is calibrated to real-world units because the viewer reads this tag. A measurement on a JPEG has no calibration — you are measuring pixels, not millimetres.

Patient Identity and Study Linking

DICOM files carry unique identifiers (Study Instance UID, Series Instance UID, SOP Instance UID) that unambiguously link every image to a specific patient, study, and series. These identifiers are the foundation of the Patient → Study → Series → Instance hierarchy that makes PACS systems work.

A JPEG file has no such linking. If the filename is lost, renamed, or the file is moved out of its folder, the association between image and patient is broken. In a clinical environment handling thousands of images daily, this is a patient safety issue: the wrong image attributed to the wrong patient can lead to incorrect diagnosis and treatment.

DICOM's unique identifiers are also what enable PACS systems to automatically organise incoming images by patient and study, what allows viewers to display prior studies for comparison, and what permits AI systems to correctly associate multiple series from the same examination.

Lossy Compression: Acceptable for Photos, Not for Diagnosis

JPEG uses lossy compression — the algorithm permanently discards information deemed visually unimportant to reduce file size. For a holiday photograph, the compression artifacts are invisible. For a radiograph, they can obscure fine trabecular detail, create false edge enhancement that mimics fractures, or smooth out subtle periosteal reactions.

DICOM supports both lossless compression (JPEG 2000 lossless, JPEG-LS, RLE) and uncompressed storage, preserving every pixel value exactly as acquired. When lossless compression is applied, file sizes are reduced by 30–50% without any information loss. This is why regulatory bodies and radiology guidelines consistently recommend against diagnostic interpretation from lossy-compressed images.

The AI Dimension

For AI-based analysis, the format difference becomes even more consequential. Machine learning models for OA screening and KL grading are trained on DICOM data with standardised preprocessing (windowing, normalisation, spatial calibration). When a JPEG is fed to these models instead, three problems arise: the bit-depth reduction may eliminate the subtle features the model relies on; the absence of pixel spacing metadata means any spatial measurements are uncalibrated; and JPEG compression artifacts introduce noise that was not present in the training data.

The Practical Reality in 2026

For years, the complexity of DICOM was a practical barrier — viewing required specialised software installed on dedicated workstations. This is no longer the case. Modern web-based DICOM viewers render full-fidelity DICOM data directly in the browser, with measurement tools that leverage pixel spacing metadata for calibrated clinical measurements.

The Salnus Surgeon Portal is one example of this approach: drag-and-drop DICOM upload, GPU-accelerated rendering, calibrated measurement tools, and AI-assisted analysis — all running client-side in the browser with no data transmitted to external servers. The days when "DICOM is too complicated" was a valid objection are over.

When you receive referral imaging, insist on DICOM. The diagnostic and clinical value difference is not incremental — it is categorical.

---

Disclaimer: This article is for educational purposes only. Clinical decisions should be made by qualified healthcare professionals.