The "doctor's doctor."

Pathologists rarely meet the patient — but their report drives what happens next. When a surgeon removes a tumor, the pathologist determines whether it's cancer, what type, how aggressive, and whether it was fully removed. Oncologists can't choose a therapy until that report lands.

Two great branches.

Molecular & computational pathology cuts across both — and it's where cancer research is moving fastest.

Where structure tells the story.

Where measurement tells the story.

From patient to diagnosis.

Every slide you'll look at in the lab traveled this exact path. When you look down the microscope, you're seeing the end of a nine-step journey.

Two different questions.

A tumor can be high-grade but early-stage, or low-grade but widespread. You need both numbers to know what you're dealing with.

Hands on the microscope.

This deck was the warm-up. In the lab, you put it to work.

• Start at low power to get your bearings, then climb to high power on anything that looks off
• Compare a normal slide against a diseased one — train your eye on the contrast
• On the real case slides: hunt for the five features from slide 13 — pleomorphism, dark crowded nuclei, lost architecture

Every project starts here.

Before you dock a molecule, run a screen, or design a study, you have to know what's already been done — what's settled, what's contested, and where the gap is that your work could fill. A literature review is how you find the edge of human knowledge before trying to push it.

Narrative vs. systematic.

Start narrative to learn a field. Go systematic when you need an answer you can defend.

A fuzzy question gets fuzzy results.

"Lung cancer treatment" returns half a million papers. Narrow it until it's answerable. In clinical research the PICO frame helps structure it:

Those four pieces become your search terms. The question writes the query.

One concept, every synonym.

Authors might write "heart attack," "myocardial infarction," or "MI." MeSH — Medical Subject Headings — is a standardized vocabulary where all of those map to a single official term. Search the MeSH term and you catch every paper on the concept, no matter the wording.

MeSH also has a hierarchy — searching a broad term can automatically include its narrower children ("explode").

Build the query like a circuit.

Trim the pile, then widen the net.

Citation chasing: found one great paper? Read its references (backward) and see what cited it (forward). Often faster than any search.

You will not read them all.

A good search returns dozens to hundreds of hits. Triage in layers — spend seconds before you spend hours.

Decide your inclusion/exclusion criteria up front (years, study type, language) so you're not deciding case-by-case and drifting.

The three-pass method.

Never read a paper front to back on the first go. Make three increasingly deep passes — most papers get cut after pass one.

Read figures first. In a good paper the figures tell the whole story — often faster and more honestly than the prose.

Not all published science is good science.

Peer review is a filter, not a guarantee. Read with these questions live — the same critical eye behind CRS's meta-research work on reporting quality.

A system beats your memory.

Capture the citation the moment you decide to keep a paper. Re-finding a lost source is the single biggest time sink in any review.

Synthesis, not a book report.

A review isn't a list of summaries ("Smith found X. Jones found Y."). It's an argument organized by theme: here's what the field agrees on, here's where it conflicts, here's the gap — and here's where my work fits.

On AI tools: fine for finding and summarizing — never trust a citation you haven't opened yourself. Models invent plausible-looking references. Verify every one.