Every article belongs to one of six pillars. Pick the one that matches where you are — each has a page explaining what it covers and listing the published articles with brief context.
Evidence-reviewed. Citations throughout. A paper lands in your inbox. The headline: a deep learning algorithm achieves 94% accuracy in fracture detection, outperforming both radiologists and orthopaedic surgeons. A colleague forwards it with a one-line message: “thoughts?” Whether your reaction is excitement, scepticism, or indifference, the answer depends on understanding what kind of system this actually … Read more
A deep learning tool matches Dixon imaging for paraspinal fat fraction quantification from standard MRI. The more clinically important finding: most of the relationship between muscle degeneration and disability operates through muscle endurance, not structural change directly.
Fracture healing assessment on serial X-rays is more subjective than it looks. A deep learning framework using YOLOv11 tracks callus formation and fracture-line changes across follow-up, with its clearest advantage at early postoperative stages when human reader agreement is lowest.
A validated ML tool estimates 1-year mortality risk at first presentation in elderly proximal humerus fracture patients, using data already available at the time of the injury. The output is for the consent conversation, not the operative decision.
ChatGPT-5 reproduces AO Spine DCM guidelines accurately for clinicians. For patients, it loses critical clinical details and consistently exceeds recommended reading levels — even when explicitly asked to simplify. This has implications for consent.
A 200-case evaluation of four LLMs on real spine surgery scenarios, scored by five spine surgeons, found significant performance differences between models. Gemini 2.5 Flash outperformed GPT-4o. Model choice is not cosmetic.
LLMs can apply OTA/AO classification to fracture radiology reports reliably at the broad level — but subgroup accuracy fails where it matters, and hallucinations occur in documented, specific ways. Here is what that means in practice.
Before you read anything about the algorithm in this paper, consider the number it was built around: 31.1%. That is the proportion of elderly patients who were completely opioid-naïve before their hip fracture and were still collecting opioid prescriptions a year later. Nearly one in three. In a cohort of more than 26,000 patients drawn … Read more
This is the Trenowin Digest for May 2026. It is preserved here as a permanent record. For the current edition, follow this link to the digest page. Monthly Digest — Archive The Trenowin Digest May 2026 The May 2026 edition of the Trenowin Digest — a round-up of what was published, what was interesting, and … Read more
Practical. No hype. I have eleven sessions on my calendar this week. Claude added all of them from a spreadsheet in under two minutes. I did not open the calendar once. That’s not a demo. That’s Tuesday morning. The rota comes through, I pass it to Claude, and the week is in my calendar before … Read more
Practical. No hype. The most enjoyable thing I’ve built with AI this year has nothing to do with surgery. It’s a debate council — six historical figures convened around a table to argue any question I throw at them. Socrates asks questions that destabilise everyone else’s answers. Marcus Aurelius finds equanimity in the cosmic indifference … Read more
Practical. No hype. The most common mistake I see surgeons make with AI is using Chat for everything. Chat is fine. It’s where most people start and it’s a genuinely useful tool. But it’s one mode of one product, and using it for tasks that other modes handle better is like doing everything with a … Read more
Practical. No hype. Fifteen minutes. That’s genuinely all it takes to go from an account that produces generic answers to a setup that knows your role, your writing style, your patients, and how you prefer to work. Most people never do this. Most people get generic answers. This post is the setup guide. Do this … Read more
Evidence-reviewed. Citations throughout. A paper lands in your inbox. A machine learning model for distinguishing spinal tuberculosis from pyogenic vertebral infection and spinal metastasis on MRI. Accuracy: 98.3%. The abstract is confident. The supplementary figures are polished. Your procurement team is interested. Before you endorse it, do some arithmetic. This paper (PMID 42082966) is an … Read more
Practical. No hype. You spend twenty minutes giving Claude the context it needs — your role, your department, what you’re working on, how you prefer to communicate. It gives you exactly what you need. You close the tab. Next time you open it, the slate is blank. Claude knows nothing about you. You start again. … Read more
Evidence-reviewed. Citations throughout. The most dangerous thing about a wrong answer is not that it is wrong. It is that it sounds confident. In orthopaedic practice, this problem has a name everyone recognises: the missed scaphoid. A normal-looking X-ray. An unremarkable report. A patient who returns six weeks later with avascular necrosis and a question … Read more
Practical. No hype. The first time I asked Claude to write a referral letter, it came back polished, professional, and completely wrong. Not factually — the clinical content was fine. But it didn’t sound like me. It sounded like someone who had read a lot of referral letters and written a composite. That’s not what … Read more
A closer look at the evidence. I used ChatGPT for about eight months before I switched. I’m not evangelical about it — plenty of people use ChatGPT well — but there were a few things that happened in quick succession that made me want to reassess, and once I looked properly, I didn’t go back. … Read more
Practical. No hype. Most surgeons who ask me about AI assume the expensive tier is the right starting point. It usually isn’t. I made the same mistake: I paid for the top option before I understood what I actually needed. This post is the answer I wish I’d had before I spent any money. The … Read more
Evidence-reviewed. Citations throughout. Degenerative cervical myelopathy is the most common cause of non-traumatic spinal cord injury in adults and one of the most consequential diagnoses to miss. The pathology is progressive, the window for intervention matters, and the early signs on MRI — subtle cord signal change, mild compression at a single level — are … Read more
Evidence-reviewed. Citations throughout. A patient presents to the emergency department with a spreading soft tissue infection and fever. The orthopaedic team is asked to review. Inflammatory markers are elevated, the limb is swollen and tender, and the clinical picture is consistent with serious deep infection — but the critical question is not yet answered: is … Read more
Evidence-reviewed. Citations throughout. When a fracture is missed on an AI-assisted radiograph review, who is responsible? The answer, under current UK law and NHS governance frameworks, is clear: the clinician. But the question is worth dwelling on — not because the answer will change soon, but because the implications of that answer are not yet … Read more
A closer look at the evidence. A patient sits in front of you for a hip replacement consultation. You’ve reviewed the X-rays, taken a history, and confirmed the indication. Before you write in the notes, an algorithm has already estimated their 90-day complication risk, flagged potential issues with bone stock, and generated a templating plan … Read more
Practical. No hype. You are preparing for your ST6 interview. The question about AI is coming — it comes up in almost every panel now. The problem is not that trainees don’t know anything about AI. It’s that most are not sure which parts of what they know actually matter. This post is the answer … Read more
A closer look at the evidence. The first time I scrubbed to a robotic arthroplasty list, I made the same assumption most trainees make. I assumed the technology would do something recognisable — that I would watch it operate, and learn from watching. What I actually learned is that robotic surgery doesn’t work that way, … Read more
Evidence-reviewed. Citations throughout. You are three months into a new post when the department takes delivery of a robotic system. There are two types of trainee in that room. One assumes it will make arthroplasty easier. The other wonders what it means for learning to operate. Both need the same thing: a clear account of … Read more
Evidence-reviewed. Citations throughout. The ED doctor is confident. The AI has cleared the radiograph. The patient has a tender anatomical snuffbox and a mechanism consistent with a scaphoid injury. The question you are answering — at 3am, on the phone — is whether you trust the algorithm. This is not a hypothetical. AI-based fracture detection … Read more