Despite treatment, patients’ cancer can get worse. Knowing when patients’ cancer got worse allows researchers to assess and improve cancer treatment. Right now, trained medical staff must look through patients’ records to find this information. This prevents them from doing other important duties, costing the health care service time and money.

This is a “proof of principle” plain text explanation!

AI Analysis of Voice to Aid Laryngeal Cancer Diagnosis

Surgical Skill Improvements Using AI-Driven Improvements in Low-Resource Settings

Victoria is a PhD student focusing on the early detection of upper GI cancers from routine blood tests measured in UK primary care data.

Vital signs like heart rate, blood pressure, and oxygen levels are important for monitoring patients in hospital. These are used in scoring systems to predict when a patient might be at risk of dying. However, current systems don’t consider how these signs change over time, which means they often trigger alerts after a patient has already significantly deteriorated. We have used AI to detect patterns in these vital sign trends using NHS hospital data. We also found that the hospital a model is trained on can affect how it makes predictions. Overall, these AI models were better at spotting patients who were likely to die, and they did so earlier than current early warning systems.

Abdominal surgery is the most common treatment for many types of abdominal cancers. After abdominal surgery, very few patients die but a large number have poor outcomes like infections. As a result, many patients need to return to hospital after being sent home.

Jack finished as a PhD student in 2025, having completed the thesis “Artificial intelligence for ovarian cancer diagnosis from digital pathology slides”. This work focused on supporting pathologists to accurately diagnose specific types of ovarian cancer from surgical tissue samples.

About me

I’m improving treatment for patients who have cancer, and are being treated with a certain type of ultrasound. This involves the same type of technology as baby scanning but at a higher pitch. This allows the beam to heat up the cancer and kill it. The planning for this type of treatment is really difficult and takes a long time. Therefore, I want to use AI to help plan better treatment.

Over 40,000 people are diagnosed with colorectal cancer each year in the UK, with a worrying rise amongst younger adults. For patients in England diagnosed with metastatic colorectal cancer, the 5-year net survival rate is only ~10%. For metastatic patients, there are only a few drugs recommended for anti-cancer treatment. High amounts of certain proteins can predict response to these treatments. Single molecule localisation microscopy (SMLM), a relatively new technique, can locate individual proteins in cancer samples.

In my project I will image colorectal cancer samples using SMLM. From these images, I will then see if we can better predict response to treatment using the organisation of these proteins. This will use both machine learning and more traditional models.

Knee osteoarthritis is a common problem where your knee becomes stiff and painful, stopping people being able to do daily tasks such as climbing stairs. No one really knows what causes it, and there are no good treatments. Predicting who might develop osteoarthritis in the future would let us step in and prevent the disease before it causes pain.

Heart failure is often diagnosed too late and has a high death rate—over 50% of patients die within five years of diagnosis. Currently, it’s detected using a finger-prick blood test, but this test usually only shows results when the disease has already progressed. Not all patients get this test early enough. To help catch heart failure earlier, researchers are using deep learning to analyze patients’ medical records from GPs and hospitals. These records hold a timeline of a patient’s health events, and the order of these events is important. By spotting patterns in this data, the goal is to find people who are at risk and recommend early testing before symptoms become serious.

Society is unfair. Women are less likely to be given pain relief in hospital, less likely to be diagnosed and more likely to not be believed. When AI is used it can take these problems and make them worse. But it doesn’t have to! I am testing AI to check if it is being unfair, and creating new ways to check if it is being unfair. People need to be able to trust the AI before it can be used. Testing the AI to make sure it is not unfair is one important step in helping people trust it.