How to pick and finish a radiology thesis

Why do a thesis at all?

Because it makes you think like a doctor and a scientist at the same time. A thesis forces you to:

• Turn a clinical question into one you can measure.
• Design a method that gives a trustworthy answer.
• Learn how to collect and interpret data without fooling yourself.

It is not about novelty for novelty’s sake. It is about learning a process. If you learn the process well, you can repeat it any time.

What makes a good thesis topic? The three rules

1. Feasible — you can finish it with resources and time available. Retrospective designs are often easiest.
2. Focused — answer one clear question. Don’t try to do everything.
3. Valuable — it should matter to clinicians or add a measurable diagnostic or prognostic insight.

How to choose (practical checklist)

1. Scan for cases you actually see daily. If your hospital does a lot of head trauma, a TBI DTI study is better than a rare PET tracer project.
2. Match modality to skill. If you are comfortable with ultrasound, pick a sonography project. If not, you will waste time learning new techniques.
3. Check data availability. If you need histopathology as a gold standard, confirm that reports are retrievable.
4. Talk to the potential guide and ask directly: have you supervised this type before? How many similar projects have you seen succeed?
5. Estimate time. Simple retrospective imaging vs HPE correlation usually needs months, not years.

Turn a topic into a one-sentence question

Example topic from Set 1: “Diffusion tensor imaging metrics as predictors of 6-month outcome after moderate-severe traumatic brain injury.”

One-sentence question:
In adults with moderate to severe traumatic brain injury, do fractional anisotropy values in the corticospinal tract on admission DTI predict modified Rankin Scale at 6 months?

If you can write this one sentence, you already understand your study’s core.

Minimal protocol blueprint — the parts that matter

1. Objective (single primary objective).
2. Design (retrospective cohort, prospective cohort, case-control).
3. Population (inclusion/exclusion; dates; single center).
4. Index test(s) (what you measure on imaging and how).
5. Reference standard / outcome (histology, clinical score, surgery, mortality).
6. Sample size (rough estimate — even an approximate calculation will guide feasibility).
7. Analysis plan (primary statistic, confounder adjustment).
8. Timeline (realistic months to finish each step).

Example mini-protocol (DTI in TBI)

• Objective: To test whether admission FA in ipsilateral corticospinal tract predicts 6-month mRS.
• Design: Retrospective cohort, single center, patients admitted 2019–2023.
• Population: Adults 18–65 with GCS 3–12, admission MRI with DTI within 7 days, follow-up mRS available at 6 months. Exclude prior stroke, severe comorbidity.
• Measurements: FA and MD in defined ROI on CST; two independent readers; mean of both.
• Outcome: mRS dichotomized (0–3 good, 4–6 poor).
• Sample size: Pilot estimate: 80 patients to detect moderate effect (this is a start; run a formal calculation with expected effect size).
• Analysis: Logistic regression with FA as predictor, adjust for age and initial GCS. ROC curve for discriminatory performance.
• Timeline: Data collection 2 months, image processing 1 month, stats 1 month, write-up 1–2 months.

That is short, practical, and makes the work measurable.

Common problems and how to avoid them

• Too broad a question. Fix: split into primary and exploratory aims, but only power the primary one.
• Unavailable gold standard. Fix: choose a surrogate outcome that is routinely recorded. State its limitations clearly.
• Poor measurement reproducibility. Fix: define ROI and measurement protocol, do inter-rater reliability on a subset.
• Underpowered study. Fix: run a pilot and calculate effect sizes. If sample insufficient, convert to descriptive or feasibility paper. That is publishable too.

Data and stats essentials — the Feynman checklist

• Always start with a data dictionary. Know exactly what each variable means.
• For continuous imaging metrics, check distribution. If skewed, use median or log-transform.
• For primary outcome, define it clearly and stick to it. Don’t swap outcomes after seeing results.
• Use confidence intervals, not only p-values. CI tells you the plausible range for the effect.
• If you are unsure about stats, consult a statistician before analysis, not after.

Writing and defending the thesis

• Write the introduction as a short argument: what is known, gap, and how your study fills it.
• Methods should be reproducible. Assume the reader is doing your work from your text.
• In results, be honest. Negative or null results are science too if the methods were sound.
• Anticipate viva questions: limitations, bias sources, how generalizable are your results, what would you do next.

Publication mindset

Design with a target journal in mind. If data are limited, aim for a specialty journal or a methods/feasibility paper. If the results are strong, push for higher impact. Either way, publish — your thesis should leave the hospital as a paper.

Quick path to finish (30-60-90 day sprint)

• Days 0–30: finalize question, get ethical approval, build list of cases.
• Days 31–60: extract images and measures, start blinded reads, create dataset.
• Days 61–90: finalize stats, write introduction and methods. Submit to guide for edits.
  This is an optimistic plan for a retrospective study with available data. Adjust realistically for your workload.

Tools and small tricks

• Use PACS bulk export and simple spreadsheets to collect variables.
• Predefine ROI and save a screenshot template for readers.
• Use free stats tools like R or ask biostatistics department for support.
• Keep a daily log of hours spent and steps done. Progress keeps momentum.

One last thing: pick a mentor who will push and protect you

A good guide helps with feasibility, access to data, and realistic expectations. Find a mentor who responds, sets deadlines, and gives concrete feedback.

Procedure Pearls — practical viva and exam tips

• Always state one primary objective up front. Viva will ask this first.
• Memorize exact inclusion/exclusion criteria and sample size rationale. If you modified them, explain why transparently.
• Know the limitations and biases: selection, information, and confounding. Say how you mitigated them.
• Be ready to explain one key figure or table in under two minutes. That will prove you own your data.
• Practice the 90-second elevator pitch: background, question, method, key result, one sentence implication.