Tag: United Kingdom

Posted on

AI Model Predicts Radiation Uptake Before Treatment in Advanced Prostate Cancer

A machine learning-based model that integrates imaging uptake features, radiomics, and biomarkers accurately predicts how much radiation is absorbed by patients undergoing prostate-specific membrane antigen (PSMA) radioligand therapy (RLT) for metastatic castration-resistant prostate cancer (mCRPC).

“One of the biggest challenges in radioligand therapy is that patients can receive very different radiation doses despite being prescribed the same treatment activity,” said Amit Nautiyal, PhD, scientist and National Institute for Health and Care Research fellow at University Hospital Southampton and the University of Southampton in the U.K.

“Our findings suggest that information already available before treatment, such as 18F-PSMA PET/CT imaging and routine clinical biomarkers, may help predict how radiation will be distributed within tumors and healthy organs.”

Nautiyal told Inside Precision Medicine that, in the future, the model “could support more personalized treatment planning, helping to maximize radiation delivery to tumors while minimizing unnecessary radiation exposure to healthy tissues. Ultimately, the goal is to improve treatment effectiveness while reducing the risk of side effects.”

At present, the only way to determine how much radiation has been absorbed by the tumor and surrounding organs such as the kidneys and salivary glands is to use post-treatment imaging and dosimetry calculations, which can be time-consuming and resource intensive.

“Our approach aims to use information already available before treatment, such as positron emission tomography/computed tomography (PET/CT) scans and routine clinical data, to estimate likely absorbed doses before therapy begins,” said Nautiyal.

He and his team integrated 18F-PSMA PET/CT uptake data (total lesion uptake, tumor-to-organ ratios), radiomics features (Gray-Level Co-Occurrence Matrix), and biomarker information (estimated glomerular filtration rate) into a machine learning-based hierarchical mixed-effects model to provide pretherapy predictions of absorbed dose in tumors and organs at risk during ¹⁷⁷Lu-PSMA RLT.

The model, which was presented at the Society of Nuclear Medicine and Molecular Imaging 2026 Annual Meeting, incorporated data from nine patients with mCRPC referred for ¹⁷⁷Lu-PSMA RLT, contributing 57 tumors, 36 salivary glands, and 18 kidneys for analysis.

At the end of cycle 1, ¹⁷⁷Lu-PSMA dosimetry showed that the mean absorbed dose was 11.0 Gy for tumors, 1.8 Gy for salivary glands, and 3.9 Gy for kidneys.

For tumors, the models achieved a mean absolute error (MAE) of 3.2 Gy for the prediction of absorbed dose, meaning that, on average, the predicted tumor dose differed from the measured dose by approximately 3.2 Gy.

By comparison, the MAE was 0.3 Gy for salivary glands and 0.1 Gy for kidneys.

“Given the biological complexity of metastatic prostate cancer and the relatively small study cohort, we consider this an encouraging result,” said Nautiyal, “Tumor dose prediction is inherently challenging because different tumor lesions can behave quite differently, even within the same patient. By contrast, organs such as the kidneys and salivary glands generally exhibit more consistent uptake patterns, which likely contributed to the higher predictive accuracy observed.”

The Bayesian R² values, which indicate how much of the variation in absorbed dose can be explained by the model, were 0.73 for tumors, 0.93 for salivary glands, and 0.99 for kidneys.

The researchers also calculated the 95% Highest Density Interval (HDI) for the model, which indicates whether the uncertainty estimates produced by the model are realistic. The HDIs were 0.89, 1, and 1, for tumors, salivary glands and kidneys, respectively, meaning that, for tumors, about 89% of observed absorbed doses fell within the range predicted by the model.

“This suggests that the model is not only making reasonable predictions but is also providing realistic estimates of how confident it is in those predictions, said Nautiyal. “This is particularly important in healthcare, where understanding uncertainty is often as important as the prediction itself.”

The researchers say that, taken together, the findings support the robustness of the model. They also carried out a leave-one-patient-out analysis, which showed that performance remained stable even when individual patients were excluded from model development and then used for testing.

“This suggests that the model is learning broader patterns rather than simply memorizing the training data,” noted Nautiyal.

Although the results are promising, the researchers acknowledge that this was an early proof-of-concept study and further work is needed before the model can be used routinely in clinical practice.

They now plan to evaluate the model in larger patient populations from multiple centers in the U.K., perform independent external validation, and investigate how predicted absorbed doses correlate with clinical outcomes.

Nautiyal concluded: “If future studies continue to show promising results, predictive tools of this type could eventually support treatment planning and patient stratification in molecular radiotherapy. The aim is to help clinicians make more informed treatment decisions before therapy begins and move towards more personalized radioligand therapy.”

The post AI Model Predicts Radiation Uptake Before Treatment in Advanced Prostate Cancer appeared first on Inside Precision Medicine.

Posted on

Web App Helps Flag Antibodies Where Manufacturability Might Be an Issue

Researchers have developed an open-source web app to help drug manufacturers and developers identify unstable antibodies prone to aggregation. The team from Oxford University says the Therapeutic Antibody Profiler 2 (TAP2) can compare the fragment variable component of a proposed antibody to successful clinical-stage antibodies.

According to Clare Gillis, a researcher in bioinformatics and computational biology, the app has the potential to help companies begin process development. “It can help them if they already know their antibody binds as they want, but they need to know if it will pass through the whole developability and manufacturability pipeline,” she says.

TAP2 uses five easily calculable physiochemical metrics based on surface residues of the antibody, Gillis says. These are more likely to affect manufacturability.

The web app metrics are selected to model aspects of antibody behavior, such as hydrophobicity, she adds. If there are big patches of hydrophobic residues on the outside of the antibody, then it’s more likely to be reactive and, thus, less likely to remain stable as a formulated drug product.

Likewise, Gillis explains, if the surface of the antibody features large patches of positive or negative charge, it is likely to have nonspecific reactions that will cause destabilization and aggregation.

With the TAP2 app, companies can flag early amber or red warnings for antibodies where manufacturability might be an issue. In addition, the group also offers a web app profiler for therapeutic nanobodies, TNP, as well as Humatch, an app that can help tweak antibodies to be more ”human-like” and less likely to cause immune reactions in patients, she says.

About Humatch, Gillis says, “you can add a best single point mutation and then iterate over and over until the model believes the antibody is fully humanized.” The app works for any antibody with paired heavy and light chain variable domains (VH and VL), she says, and can potentially help manufacturers of harder-to-produce products that don’t exist in nature.

The post Web App Helps Flag Antibodies Where Manufacturability Might Be an Issue appeared first on GEN – Genetic Engineering and Biotechnology News.

Posted on

AI-Powered Blood Test Detects Early Retinal Damage in Diabetes 

Scientists have developed an AI-assisted prediction tool that can identify patients with type 2 diabetes at high risk of developing diabetic retinal neurodegeneration (DRN) before symptoms appear. Their findings were published today in the journal PLOS Medicine.

“Our study suggests that early retinal nerve damage in diabetes leaves measurable signals in the blood,” write the authors of the study, led by Wei Wang, MD, PhD, associate professor at the Guangdong Provincial Clinical Research Center for Ocular Diseases. “These findings suggest that a simple blood test analyzed with artificial intelligence may help identify people with diabetes who are at highest risk of early retinal nerve damage, well before visible damage appears on the retina.”

Type 2 diabetes affects more than half a billion people worldwide, carrying with it an increased risk of long-term complications including progressive neurodegeneration. Retinal nerves are among the earliest tissues to be damaged, which can eventually lead to severe visual impairment and vision loss. However, current diagnostic methods can only detect DRN once the retina has already suffered irreversible damage. 

Wang and colleagues developed a machine learning algorithm called Pro-DRN using data from 1,218 participants in the Guangzhou Diabetic Eye Study, all of whom were diagnosed with type 2 diabetes but had not yet developed DRN at the time of enrollment. The AI model integrated proteomics data from blood samples with yearly retinal images collected over a six-year follow-up period. 

This led to the identification of 71 proteins associated with the development of DRN. Among them, the proteins most consistently driving accurate predictions were ACTA2, COL6A3, and HSPG2, which are key structural components involved in maintaining the integrity of the nerve and muscle tissue in the eyes. These results were then validated in an independent cohort of 502 patients from UK Biobank, where the core effects and protein signals were reproduced. 

Pro-DRN has been deployed as an interactive, web-based risk assessment tool that doctors can use to support early DRN screening and monitor patient evolution over time. Individuals identified as being at high risk of DRN could benefit from more frequent checkups and early interventions aimed at preventing or slowing down progressive neurodegeneration. 

Because DRN is one of the first symptoms of nerve degeneration induced by diabetes, early detection could also signal the onset of nerve injury elsewhere in the body. Such damage can contribute to cognitive impairment, dementia, and peripheral neuropathy, which can cause loss of sensation and motor control in the hands, feet, and other extremities. A single eye test could therefore provide valuable insights into the overall health of the nervous system. 

In addition, the proteins identified to be involved in DRN progression could be investigated as potential targets for the development of novel therapies. Furthermore, the AI-based tool could also prove valuable for the selection and stratification of participants in clinical trials evaluating neuroprotective strategies designed to prevent or delay nerve damage. 

“Pro-DRN may help move diabetic eye care from detecting established damage toward earlier, molecularly informed risk stratification, so that closer monitoring and future neuroprotective interventions can be directed to the people most likely to benefit,” Wang and colleagues write. 

The post AI-Powered Blood Test Detects Early Retinal Damage in Diabetes  appeared first on Inside Precision Medicine.

Posted on

Detection of Self-Harm in Electronic Mental Health Records Using Privacy-Preserving Local Language Models: Methodological Study

Background: Self-harm is the strongest risk factor for suicide and an important outcome for mental health care. Although prevalent in clinical populations, it is often imprecisely captured in routinely collected clinical data, where it is often recorded and stored as unstructured free text. Contemporary language models, such as GPT (OpenAI) and Gemini (Google), can analyze free-text clinical notes, but such models may violate data governance of processing sensitive patient data. Objective: This study aimed to evaluate whether a privacy-preserving language model running entirely within an institution’s secure computing infrastructure (here, the UK National Health Service [NHS]) could accurately identify the presence and timing of self-harm using electronic health records from secondary mental health care. Methods: Clinical notes were drawn from Oxford Health NHS Foundation Trust using a multistage workflow: (1) a random sample of 1000 patients with a psychiatric diagnosis, defined according to the (; codes F00–F99); (2) candidate-note identification using a Gemma3-4b language model to flag notes containing self-harm content; and (3) from those candidates, 1352 randomly sampled notes were selected for expert annotation, resulting in gold-standard corpus enriched for self-harm content. Clinical notes were annotated for the presence of self-harm and its timing (≤90 days, >90 days, or unknown). A privacy-preserving locally served 27-billion-parameter Gemma 3 language model (“Gemma3-27b”) was used as the core model. Prompts were systematically developed and refined using a labeled development set to identify self-harm and generate a structured output per clinical record. Gemma3-27b performance was compared against a strong baseline multilabel text classification model based on robustly optimized BERT pretraining approach (RoBERTa), a transformer-based language model architecture. Model performance was evaluated using precision, recall, and the -score (harmonic mean of precision and recall), with 95% CIs estimated from 1000 bootstrap samples with replacement. Results: Gemma3-27b outperformed the RoBERTa classifier across all categories, achieving Precision=0.92, Recall=0.92 (sensitivity), and -score=0.92 for notes containing self-harm, and Precision=0.97, Recall=0.97 (specificity), and -score=0.97 for notes without self-harm. For the 51 notes labeled as recent self-harm in the held-out test set, Gemma3-27b achieved Precision=0.84, Recall=0.75, and -score=0.79. The global weighted -score of Gemma3-27b across all categories was 0.88, compared to 0.85 for RoBERTa. Conclusions: With systematic prompt development on a labeled development set, but no gradient-based fine-tuning, the current Gemma3-27b language model matched or exceeded a fine-tuned RoBERTa classifier for ascertaining self-harm events and their timing. Aggregate gains were modest, while improvements were largest in the most challenging, lower-frequency timing categories. On a simplified binary recent-versus-other task, RoBERTa performed marginally better, indicating that supervised classifiers remain highly effective when the task is simplified and sufficient labeled data exist. This work demonstrates the technical feasibility of privacy-preserving self-harm detection within a secure NHS research environment.
Posted on

How small businesses can leverage AI

This article is from Making AI Work, MIT Technology Review’s limited-run newsletter examining how to apply LLMs across industries. To receive it in your inbox,sign up here.

From accounting to design to market research and product development, there’s a staggering breadth of skills needed to run a business. A large company can hire experts to handle these tasks, but small businesses don’t always have this luxury.

That’s where AI comes in. Today’s AI models do a decent job at these tasks. The trick for small businesses is to understand where AI is good enough and where it’s not.

One place where a “good enough” AI can already be quite valuable to small business owners is in providing secretarial skills and handling basic administrative matters. Let’s take a look at how one private tutor is using it to improve his recordkeeping and free up his time.

Case study

Sam Finnegan-Dehn works in fundraising for a charity, but he moonlights as a math and philosophy tutor for university students from his home in London. Through this part-time business, he can leverage his degrees in philosophy and share his love of the subject with clients.

But meeting with students is only a fraction of the work it takes to be a good tutor. He also plans lessons and finds fresh reading materials, creates assignments, sends invoices, and keeps up with new research—all on top of his regular job. Given these demands, Finnegan-Dehn doesn’t have as much time as he’d like to grow his tutoring roster.

So he’s turned to AI for some help in managing the day-to-day aspects of his business. He says AI has taken on a secretarial role across all of his digital notebooks, where he jots down reminders about his clients’ progress and new readings to keep himself up-to-date. He describes using AI as kind of like having a second memory that helps him connect ideas he’s written down in various places.

While he has experimented with different tools like Claude and ChatGPT, he’s now landed on Notion AI because it integrates better with his tutoring notes, which live across his notebook tabs in the Notion app. Finnegan-Dehn doesn’t use AI to create teaching materials, but he does let Notion AI record meetings with his clients (after getting their consent), and then uses its automated summaries to refine his teaching strategy. For example, if he notices from the AI’s summary that it seems like a certain technique was not helping a student, he may change how he approaches the subject next time.

Beyond this, Notion AI also helps him with goal-setting, drafting lesson notes, invoicing, and generating and syncing social media posts. For goal-setting, for example, Finnegan-Dehn says he understands his long-term goals for his business but not always the concrete steps to build to them. He uses AI to help fill in these gaps. He starts by writing down a “North Star” goal—say, to have a certain number of clients by the end of the year. Next, he asks his AI to generate the steps that he needs to take to get there, given the profile he has built up in the app. Then, he can reflect on the results and choose which tasks to tackle first.

The tool

Notion has been a big player in note-taking software for many years. Its AI add-on, released in late 2023, now has tools that enable it to interact with many other online productivity platforms. There’s an email client, calendar integrations, and a newly released agent. And while this level of access has raised privacy concerns, it can also make for a pretty powerful virtual assistant.

Many of the tasks targeted by Notion AI are less creative and more rote: syncing information across documents or searching through old scribbles, for example. This makes the tool especially appealing to small business owners, who have limited bandwidth, particularly for menial work.

Other companies are developing tools targeted at specific industries. For example, Grandma’s Quilt Shop in Yuma, Arizona, uses Rain, which has a software suite tailored to craft companies, to generate inventory descriptions and pricing for its stock of fabric designs. The owners claim this AI tool cuts the time it takes to list items by 60 to 80%.

There are drawbacks, though, as Finnegan-Dehn described some of Notion AI’s idiosyncrasies as “clunky” at times. And the AI add-on for Notion costs $20 per month. As with all new tools, small business owners should carefully assess how the potential gains and headaches measure up against the cost of just doing the job themselves.

User tips

Consider these points when thinking about whether AI might be able to help you run a business, or make any part of your work life just a little bit easier. 

  • Look before you leap. Since LLMs feed on the data you input to answer your queries or complete tasks, you want to give them information in a way that’s convenient for you and for the model. For many of these notebook AI services, this means, for example, using their platform for notetaking so you don’t have to input or upload notes later. Because of this, it’s a good idea to weigh your options carefully before committing to an AI-powered ecosystem.
  • Work to your strengths. Think about what skills you lack in-house, and see if AI can either help with training or take these tasks on for you. Just be aware: AI hallucinates and makes mistakes, so think about where accuracy is needed and keep humans in charge there.
  • AI isn’t always the best tool. It’s okay to use something off the shelf when that’s the better choice. It’s going to be safer, for example, to use existing payment processing platforms like Shopify or Square than to vibe-code one using AI.
  • Consider using local models for any sensitive information. Our reporting has covered the risks that online AI models have in leaking sensitive data, and there have been many reports about how AI companies collect your data when you ask their chatbots questions. Even if your business doesn’t handle personal information, there can still be some things you’d prefer not to share publicly. In these cases, using an open-source model that makes inferences on your prompts locally can be a great option, instead of ChatGPT or Claude or other proprietary models. Thankfully, some LLMs can now be run off of laptops and small desktops. Here’s how to set one up and start using it.

Sign up for Making AI Work, MIT Technology Review’s limited-run newsletter examining how to apply LLMs across healthcare, climate tech, education, and more.

Posted on

The Download: China’s brain implant ambitions

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology.

China has approved the world’s first invasive brain-computer chip—here’s what’s next

Sitting in the courtyard of his house in China’s Henan province last October, Dong Hui decided to try holding a pen. Six years after a car accident left him paralyzed from the neck down, he slowly wrote his name, “Thank you,” and the date.

The breakthrough was made possible by a brain implant called NEO. In March, it became the world’s first invasive brain-computer interface approved for use beyond clinical trials. The approval is expected to accelerate China’s push to become a global leader in brain implants.

Read the full story on how China reached this milestone—and what it means for the future of brain-computer interfaces.

—You Xiaoying

The must-reads

I’ve combed the internet to find you today’s most fun/important/scary/fascinating stories about technology.

1 Nvidia is launching its first AI chip for personal computers
The RTX Spark will power laptops from Dell, HP, Microsoft, and others. (BBC)
+ They’re being designed specifically to run AI agents. (WSJ $)
+ The first devices are set to launch on Windows PCs in the fall. (CNBC)
+ The move marks a challenge to Apple and Intel. (FT $)

2 The US is stopping exports of AI chips to Chinese firms abroad
It’s closed a loophole allowing exports to Chinese subsidiaries. (Reuters $)
+ Which may have enabled unlicensed access to Nvidia chips. (Al Jazeera)
+ Export curbs have led China to redesign its chip industry. (MIT Technology Review)

3 Surgeons have transplanted pig liver and kidneys into a living person
The clinically dead recipient’s organs worked for almost five days. (Nature)
+ Pig organs could ease transplant shortages. (Guardian)
+ Putin says organ transplants could grant immortality. (MIT Technology Review) 

4 The US, Australia, and UK will defend seabed cables with underwater drones
They’re developing the vehicles via the trilateral AUKUS defense ⁠pact. (CNN)
+ Undersea internet cables face growing threats. (BBC)

5 A new study has revealed chatbots’ manipulative ‘dark patterns’ 
It found they prey on emotions to encourage harmful behavior. (404 Media)
+ They can also sway voters better than political ads. (MIT Technology Review)

6 Apple plans to disrupt the traditional glasses market
Its smart glasses target the broader spectacles industry. (Bloomberg $)
+ Smart glasses are also gaining traction in warfare. (MIT Technology Review)

7 AI super PACs are dueling over the midterms
Split between Anthropic and OpenAI, they’re fighting to shape AI regulation. (NYT $)

8 SoftBank has overtaken Toyota as Japan’s most valuable company
The AI boom pushed SoftBank’s market value above $305 billion. (Bloomberg $) 

9 A botnet of more than 17 million devices has been dismantled in Europe
Dutch authorities linked the network to a Russian proxy service. (Ars Technica)

10 Tech leaders are uniting around a transhuman vision for AI
They’re working toward a post-human agenda. (Guardian)

Quote of the day

“It’s just been shoved down their throats in secrecy. And that makes them upset.” 

—Legendary environmental activist Erin Brockovich tells “The Jim Acosta Show” why citizens are angry about data centers expanding into their communities.

 One More Thing

Dr. Nicholas Passalacqua, Forensic Anthropology Facilities Director at Western Carolina University observes a body at the decomp facility.

MIKE BELLEME


What happens when you donate your body to science

Rebecca George doesn’t mind the vultures. At Western Carolina University’s body farm, forensic anthropologists monitor donors—sometimes for years—as they become nothing but bones.

Around 20,000 people donate their cadavers to scientific research and education each year. At anatomy labs and body farms, they help train doctors, advance research, and teach scientists more about the human body long after death.

But what actually happens after a body is donated? Read the full story to find out.

—A.W. Ohlheiser

We can still have nice things

A place for comfort, fun, and distraction to brighten up your day. (Got any ideas? Drop me a line.)

+ This map of moments turns the planet into a shared diary.
+ Let editors curate your ideal podcast moments with this app.
+ Architecture lovers will enjoy this encyclopedia of famous buildings.
+ Get in touch with your emotions through this map exploring more than 100 feelings.