In many scientific fields, we measure things by comparing them to a baseline. To understand if a heart is healthy, we compare one wall to another; to see if a brain region is inflamed, we compare it to the cerebellum; to normalize genomic data, we compare a probe's intensity to the average of the slide.
This approach, known as relative normalization, works perfectly when pathology is localized. But as a growing body of research shows, it creates a dangerous "blind spot" when a condition affects the entire system uniformly. When the baseline itself is diseased, the relative difference disappears, making a severely ill patient appear perfectly normal.
The Trap of Relative Normalization
The intuition behind relative normalization is simple: by comparing a target region to a "healthy" reference within the same subject, we can cancel out noise from scanner settings, patient movement, or technical artifacts.
However, Jirou Sasaki and colleagues demonstrated as early as 1990 that this logic fails in the presence of global impairment. In medical imaging, if a disease affects an entire organ uniformly, there is no "healthy" reference point left. The software, looking for a difference that isn't there, may incorrectly identify the state as normal by normalizing against a compromised reference.
This isn't just a theoretical risk; it is a frequent clinical reality in "balanced multivessel disease." As Marcelo F. Di Carli and Jae Ho Sohn explain, if all three major coronary arteries are narrowed to the same degree, the blood flow across the heart remains uniform. A traditional relative scan will show a "homogeneous" heart, leading clinicians to miss high-risk cardiovascular conditions because no single region stands out as a comparative deficit Sohn et al..
When the Reference Region Moves
The problem extends beyond the heart. In neuroimaging, researchers often use a "pseudo-reference" region, like the cerebellum, assuming it is unaffected by the pathology being studied.
Kuriko Kagitani‐Shimono and colleagues found that if the disease actually involves that reference region, the resulting signal-to-noise ratio is underestimated Kagitani‐Shimono et al.. Similarly, Tun Wiltgen and Mark Mühlau note that systemic neurological diseases can alter the entire brain's intensity profile. If you normalize based on internal brain tissue that is itself pathologically altered, you introduce a bias that obscures the very markers you are trying to measure Wiltgen et al..
This "moving baseline" problem also appears in statistics. Julian McGinnis and Cui Ci Voon argue that standard Z-score standardization—which scales data based on the mean and variance of the whole group—can inadvertently "wash out" biological signals. If the entire patient group is affected by a disease, the group mean shifts, masking the true difference between patients and healthy controls McGinnis et al..
Technical Artifacts Masquerading as Biology
In genomics, the failure of local normalization can even create "phantom" biological patterns. Haiyuan Yu and Jiang Qian discovered that many reported instances of gene co-expression were actually artifacts of where probes were physically located on a microarray slide.
Standard normalization techniques designed to fix these spatial biases often reduced the signal of distant data points while leaving short-range technical noise intact. This selective correction artificially enhanced the appearance of local clusters, leading researchers to believe they had found "chromosomal clustering" when they had actually just found a local chip artifact Yu and Qian.
The Solution: Absolute Physiological Units
To solve the "arbitrary units" problem, researchers are moving toward absolute quantification. Instead of asking "How does this region compare to that one?", they ask "How many milliliters of blood are flowing through this gram of tissue every minute?"
As Marc Dewey and Maria Siebes argue in a consensus statement, transitioning to absolute physiological measurements (like mL/min/g) is the only way to accurately diagnose patients with global reductions in circulation Dewey et al..
This shift is being enabled by new hardware and modeling techniques:
- Dynamic PET and SPECT: By using kinetic modeling, researchers like Simona Ben‐Haim can now quantify myocardial perfusion reserve, identifying reduced flow capacity even when the heart looks homogeneous Ben‐Haim et al..
- Automated Factor Analysis: Tools like GFADS allow for highly reproducible absolute blood flow measurements even in clinics without expensive on-site particle accelerators Kardan.
By anchoring data to physical constants rather than internal averages, researchers can finally see the "invisible" diseases that affect the whole system at once.