Amidst the media hype about Artificial Intelligence (AI)’s ability to replace physicians, we forget a fundamental truth: AI without a pathologist is merely a blind algorithm.
We often see visually impressive technical tools that fail miserably when applied in real laboratory workflows. The reason is simple and critical: the absence of clinical context.
In this article, we discuss why the pathologist’s presence at the heart of AI development is not just an “addition” but essential for a project’s survival, drawing insights from field experts like Dr. Diana Montezuma (Head of R&D at IMP Diagnostics).
1. A Lesson from the Past: Don’t Repeat the Mistakes of Viewers
Do you remember the first generation of Digital Pathology Viewers? They were catastrophic.
The reason was that they were developed by brilliant programmers who had never spent a single day behind a microscope. The result was software lacking the most basic tools a pathologist needs for daily diagnosis.
The same scenario is repeating today with AI. If the pathologist is not involved in algorithm design from “day one,” the result will be a technically impressive tool that is clinically “unusable.”
2. Pathology is Not an “Exact Science” as Programmers Assume
One of the biggest shocks data engineers face when entering our field is discovering that pathology is not 1+1=2.
The programmer sees: data, pixels, and absolute numbers.
The pathologist sees: biological variability, disease context, and subjectivity in assessment.
This “technical ambiguity” is the essence of medicine. The programmer needs a pathologist to explain that variations in cell morphology do not necessarily indicate a different pattern; they might just be an artifact. Without this guidance, algorithms will train on “noise” instead of true signals.
3. Where Does the Real Value Lie? (Not Just in Speed)
There’s a misconception that AI is only here to accelerate routine work. The truth is, pathologists are already very fast and efficient, and the cost of routine “eyeball” diagnosis is relatively low. Therefore, tools that merely attempt to “mimic” what we do quickly often fail to prove their economic viability.
The real revolution lies in “what the eye cannot see”:
Predicting treatment response (Predictive Biomarkers).
Precise quantitative analysis (Quantification) that surpasses human estimation.
Extracting genomic information from tissue images (Prognostics).
This is where we need AI: to be a “super lens” that sees what we cannot, not just an assistant performing routine tasks.
4. Ultimate Responsibility: Human First
No matter how technology evolves, the ethical and legal question remains: who bears responsibility?
An algorithm cannot be sued, nor does it hold a “license to practice.” Responsibility always rests with the human physician.
As a recent Time Magazine article noted, AI may revolutionize, but it lacks “compassion” and “complex clinical judgment” (Critical Thinking). Therefore, leadership must remain with humans, while AI acts as a smart and reliable co-pilot.
Conclusion
The message is clear for developers and for us as physicians: a multidisciplinary approach is the solution.
The future of digital pathology cannot be built in closed server rooms. Algorithms must be developed hand-in-hand with those who will use them.
If you are a pathologist, do not fear technology; instead, participate in its creation to ensure it serves your patients, rather than increasing your burdens.
