Applied AI for Medical Imaging MSc postgraduate personal statement example

Personal statement example

Once a month I sit down with the reject log for our general X-ray rooms. Every image a radiographer discards and repeats is recorded with a reason: rotation, clipped anatomy, motion, wrong exposure. Most of the entries are routine, but read together they show where the extra dose and lost minutes in our department actually come from. That habit of treating images as data, not just pictures, is what has led me to apply for postgraduate study in applied AI for medical imaging.

I graduated with a 2:1 in Diagnostic Radiography and have worked for two years as a radiographer in a district general hospital, mostly in plain film with a rotation into CT. My undergraduate dissertation was a retrospective reject analysis of adult chest radiographs taken over three months on placement. I categorised around four hundred rejected images by cause and compared rates between inpatient, outpatient and portable examinations. The clearest finding was that positioning errors, especially rotation and missed costophrenic angles, made up most repeats on portable examinations, where patients often cannot sit upright. The project taught me how much depends on clear definitions. Two of us coded a sample independently and initially disagreed on a surprising number of cases, so I rewrote the criteria with example images before finishing. I now recognise that same problem in the labelling of training data for machine learning.

At work, I see AI from the user's side. Our department has started trialling software that flags possible findings on chest radiographs for reporting staff, and radiographers were asked to give feedback on how the flags appeared in the workflow. I am not involved in evaluating its accuracy, but conversations with colleagues raised practical questions I want to understand properly: how a model trained on one hospital's equipment behaves on another's, what happens with poorly positioned images like the ones in my reject log, and who notices when performance drifts.

To prepare, I have spent the past year learning Python in the evenings through free online courses, then working through introductory material on convolutional neural networks. As a personal project I trained a simple classifier on a small subset of a public chest X-ray dataset to separate frontal from lateral views. It is a modest task, but it taught me about train and test splits, why accuracy alone can mislead on unbalanced classes, and how easily a model can pick up on text markers burnt into the image rather than anatomy. I also revisited the linear algebra and statistics I last used at A level, since I know the mathematical side of the course will be demanding.

Outside work, I ring church bells with a local band on Thursday evenings and most Sundays. Learning methods involves memorising patterns and holding your place while the sequence shifts around you, and it has made me patient with slow, cumulative skills. I also tutor my younger brother in GCSE maths, which has improved my own ability to explain ideas without jargon, something I expect to need when talking to clinicians and engineers alike.

I am applying now because I want structured training in the methods behind these tools, rather than continuing to piece them together alone. I bring an understanding of how images are acquired, why they go wrong, and how departments actually work under pressure. In return I hope to gain the technical grounding to evaluate and help build imaging software critically, and in time to work in a role between clinical teams and developers, where knowing both the patient on the table and the pixels on the screen is useful.