My guest today works at exactly that frontier. He is a postdoctoral scientist at the University of Oxford, whose work sits at the intersection of complex biomedical data, computational biology, and the use of AI in research. He also holds the title of AI Ambassador at Oxford. Dr. Badran El Shenawy, welcome to the show.
Thank you. It's an honour to be here.
Egypt is spending real money to digitise healthcare and expand universal health insurance. From where you sit — turning messy biological data into something a model can use — what's the actual difference between a system that's digitised and one that's genuinely AI-ready?
That touches on a major point around all of AI and machine learning. The model is dictated by the data input into it. A digitised document — for example, a scanned PDF — is digital, but it's not machine-readable. You cannot input that into a model. The process of transforming that scanned PDF into pixels, numbers, and structured data that models can be trained on and indexed — that's AI readiness. That is exactly the difference between digitisation and truly AI-ready data.
What is the single biggest mistake you hear when people talk about AI in healthcare?
The misconception that building the model or the algorithm is the difficult part. In reality, that's the easy 10%. Ninety percent of the work is data curation — collecting the right data, curating it, ensuring it follows the correct standards, that it's usable and consistent across different models. People assume the fancy result at the end is all that matters, when in reality it's the data that matters. There's a common saying in science: garbage in, garbage out. If your data is bad, even the best models in the world will give you poor results.
Investors in Egypt are already acquiring AI diagnostics startups. How do you tell a real AI healthcare company from one that's just bolted an AI label onto an ordinary service?
Three things. First, ask the company where their model fails. A genuine AI diagnostics company will know exactly where its model has limitations — whether due to data bias or model capabilities — and will know where human judgement is still necessary. If they can't answer that, it's a red flag. Second, are clinicians involved in development? If it's purely computer scientists and engineers without clinical input, the model may perform brilliantly on test data and never work in a real clinic. Third, has the model been validated on external datasets beyond the training data, and has it shown actual utility in a clinical setting? Those are the three criteria for identifying a genuinely AI-driven company.
If Egypt had to pick where AI in healthcare pays off first — diagnostics, hospital operations, patient records, drug discovery, public health planning — where would you start?
I'll start with what has the lowest ROI, which is drug discovery. It's the fanciest and everyone aims for it, but it's a decade-long bet at minimum. Where Egypt stands to gain most significantly is in screening, diagnostics, hospital operations, and patient records. AI can help immensely with organising patient records, with diagnostics, and with screening. When clinicians are overwhelmed and exhausted at 2am, information can easily be missed. A model will not miss that information from a diagnostic report. In that scenario, the model is a co-pilot, not an autopilot — and that is where the highest ROI lies.
Three to five years out, what would tell you Egypt has genuinely moved from digitised healthcare to investable AI-enabled healthcare — and what would tell you it has stalled?
Success looks like two things. First, full digitisation and organisation of patient records — where a clinician in Cairo can instantly pull up patient records from Minya or any other governorate. Complete transparency of information, democratised and accessible across disciplines. Second, moving from pilot studies to actual daily clinical application across major centres, gradually trickling down to smaller ones. Where it has stalled is if pilot studies remain pilots indefinitely, and if the utility of AI has not been communicated effectively to patients. Because at the end of the day, the data comes from patients. If they don't understand the process and have no faith in it, they won't provide the data. That's how you know the process has stalled.
In one sentence — are you optimistic?
Highly optimistic. The work being done in Egypt using AI models to detect breast cancer through mammograms — using our own data and our own disease patterns rather than European models — is a perfect example of the fascinating and important AI work happening here. That's close to my heart, having specialised in triple negative breast cancer during my PhD.
That's the perfect note to end on. Dr. El Shenawy, it's been a great pleasure having you with us. Thank you.
It's been an honour. Thank you so much.
