thousands of residents and tourists have been evacuated from towns north of athens, as wildfires spread across the country. more than 150 wildfires are burning out of control — some now close to the greek capital. two people are known to have died, and at least 20 others have been injured. the afghan government says american b—52 bombers have killed more than 200 taliban fighters in a series of air strikes in the northern city of sheberghan. the us and the uk have asked their citizens in afghanistan to leave the country immediately because of the worsening security situation. 0n the final day of the tokyo 0lympics, the men's marathon has been won by eliud kipchoge of kenya, the defending champion and world record holder. more than 100 athletes from a0 nations took part in the marathon in sapporo in northern japan.

wales has moved to covid alert level zero, meaning social distancing rules, and most other restrictions, have now come to an end. people can gather indoors, and nightclubs are open once again, but face masks are still required by law in shops and on public transport. here's our wales correspondent, hywel griffith. it's not quite ripping up the rules but from today, life in wales is far less limited. social distancing is no longer written into law. no need to count how many people meet indoors. facemasks are still mandatory in shops and most indoor spaces, something emma says she can live with. i think it's quite a good idea, really, because there's still quite a lot of vulnerable people. i don't like wearing them but it's just something we've all had to do, really, isn't it? for the manager here, it's about keeping customers confident. they can't afford to take risks. we need to move steadily, gently to complete normality.

it isn't going to happen overnight but we do need to build our trade up because since we've reopened, we're only taking about 60% of what we took in 2019. while there's obvious relief at ending some of the restrictions people have had to live with for almost a year and a half, there's already some nervousness about the months ahead. the welsh government has warned that the rise of another coronavirus variant could yet mean having to reimpose some control measures. barring people from the bar again would be unpalatable for many in the beer trade. from today, table service and masks are no longer legally required in welsh pubs, restau ra nts a nd cafes. they hope they've seen the last of lockdowns. basically, we've taken no money. we've managed to survive with the brewery, we've got an independent brewery. plus we've been doing home deliveries. but if we hadn't have had that, i don't know if we'd have survived and i think that pubs in general are struggling. everyone hopes today is a first

taste of real lasting freedom. but for now, few are getting carried away. in hywel griffith, bbc news, maesteg. now on bbc news, click. this week: a special programme. can artificial intelligence make healthcare better for all of us, and save the nhs? here in the uk the national health service has been in crisis for many years, looking after an increasingly ageing population with complex needs. fighting for resources, it has been at breaking point. and then the pandemic hit.

there will come a moment when no health service in the world could possibly cope. and indeed, many countries�* health systems have struggled with the sheer number of covid patients. even those which didn't have people spilling out onto the streets had to put all other treatments on hold. the nhs is an old health system. in fact, it's several different systems that sometimes work together and sometimes don't. and now, it's trying to reinvent itself and embrace technology to beat the queues that covid—i9 has created. this is a story i started filming just before the first uk lockdown, when we weren't yet wise to masks or social distancing, but there was definitely a hint that something was coming. i wanted to find out how artificial intelligence could be used to help take up

the strain that the nhs was already under. and, on what turned out to be my last day in london for many months, i found myself about to do something that i'd never done before. i wasn't nervous until i asked my twitter followers what i can expect. "very loud noise — claustrophobia." now i'm nervous. i'm going to have an mri scan on my liver. and i have to lay absolutely still? yep, and there's breathing instructions. this magnetic resonance imaging scanner — mri scanner — is going to be able to see my soft tissue and my fluid in great detail. and it will produce images like this, which will be looked at by a radiologist to try and work out if they can see anything unusual. now, the thing produces really strong magnetic fields, which means nothing metal can come in — and that includes you. but this is no ordinary mri scan. instead of being read by a human, my mri is going to be read by an artificial

intelligence. this is a computer that will look at the images and read them at least as well as a human radiologist would. breathe out and hold your breath. now, this film is not about computers stealing jobs. this is a film about computers filling the gaps in an nhs that is short of money, short of nurses, and short of the highly skilled radiologists who can read and interpret images like these. but that's a really difficult job that takes years of training. i mean, how could you teach a computer to do that? well, here's a classroom full of medical students, and this is a computer. the way you teach each is very different.

what you have in front of you here are ct scans of a human lung with cancerous tumours. so i'll be teaching you how to identify them and hopefully save lives. the human brain is brilliant at learning things. it can understand spoken words. it can understand diagrams. just through the teacher's descriptions, these students can have a good guess at finding tumours in these images. they get some right, they get some wrong. but with repetition and practice, their brains make more and more connections — which strengthen with success until eventually we really understand the task in front of us, and we can do it well. but computers are dumb.

they literally know nothing. they can't look at a diagram and imagine it in real life. they can't understand spoken words. normal teaching methods won't work. so, instead of trying to describe to them what we want them to learn, we teach them using trial and error. millions of trials, in fact. the computer starts by circling completely random parts of the image. itjust guesses — it doesn't even know what it's getting at. all it knows is when its guess is right and when it's wrong. and mostly, because it's guessing, it's going to get it wrong. and the great thing about computers is they can do this over and over again, really fast, and they remember everything. and every time it does

accidentally get it right, it makes a connection to its previous correct guesses. with every right answer, certain connections get stronger. and with every wrong answer, others get weaker. through this barrage of guesses a network grows, similar to the human brain, which starts to distinguish between right and wrong. a digital thought process that we call a neural network. until eventually, after a multitude of attempts, it has very few failures and a lot of success. ok, let's try some more difficult scans now. so if we just take your gown off... and this is the technique that could soon help to relieve

the pressure on the uk's breast screening service. almost 1000 women die from breast cancer in the uk every month. the aim is to catch the disease early when it's most treatable. but that means that more than 2 million women have their breasts scanned for potential cancer every year. my entire role is clinical diagnostics within breast. you know, it takes up a great deal of my life, it is really important, and we are kind of fighting a big fight. bernadette works at the lincolnshire breast screening service, and she is part of a shrinking workforce. for years, the nhs has been unable to train or even recruit enough radiologists, and many services now face chronic staff shortages. this is further complicated by the fact that each mammogram needs to be read independently by at least two specialist

clinicians. so we have two people reading a mammogram, because it is actually a really difficult process, 0k? what we're looking for is tiny, tiny cancers. maybe two millimetres, maybe a little tiny smudge within a breast. and we are human, and some of us miss them. so that second read is that second opportunity to pick up that tiny little smudge that may, you know, change a woman's life. so the nottingham university hospital has been trialling a new tool that may help. this is mia, an ai trained to spot breast cancer. the aim of this project is for this system to be the second reader, potentially speeding up the whole process. so mia has had a look at this mammogram as well, can we have a look

at what mia thinks? so in this particular case, mia has actually marked the area that i would also be concerned about. and it makes a callback decision itself and as you can see there mia says callback, malignancy, because it's suspicious. its makers, kheiron medical, have been developing mia for over 3 years. here in nottingham, jonathan and his team have been testing mia on the hospital's historical data sets. these other women who have been to this clinic in the past and were diagnosed by a human. the aim is to see if mia would have made the same decisions. so this time, mia hasn't actually placed any mark on the image here, and its opinion is no recall is required. actually this lady did come back for some extra tests and this well—defined mass, she had an ultrasound when she came back and this was a cyst, a perfectly harmless cyst. so i suppose the recall would be a false positive, this lady didn't actually need to come back. a false positive is when

the reader thinks that there is cancer, but on further investigation it turns out that they were wrong. and over at the cambridge breast unit, professor fiona gilbert has been conducting research to try and reduce those false positives. something that not only put a lot of strain on the system, but also a lot of strain on the patients. it is obviously causes quite a lot of distress to the woman being called back, and the majority of them turn out to be normal. it is a lot of work for us to be assessing all of these women for a relatively small number of cancers. and when a programme screens over 2 million people every year, these little percentages do matter. every year around 70,000 women are given a false positive result, and some never attend a screening again. fiona and her team have been testing an ai built by google health to see how the software compares to human radiologists.

so in the retrospective study that was done, the algorithm performed better than some of the individual radiologists, and worse than other radiologists. so when they took the average performance of all the radiologists, the algorithm would have called back fewer people. but this research project is still in its early stages, and is set to move to the next phase of development this year. in the meantime, back in nottingham, mia has finished its initial trials and will now be tested in 15 more nhs sites against ongoing cases, to make sure that it works. artificial intelligence is the next amazing transformation, so that is what excites me. and it excites me that i don't know the full potential.

we know that the cancers we are missing tend to be smaller, they tend to be more aggressive, and those are the ones that we want to find. but as the pandemic hit, all of this had to stop. and as breast clinics emptied out, elsewhere the nhs was about to be overrun. the coronavirus pushed intensive care units beyond their limit. but even before the pandemic, these wards had been running at very near capacity. and it is here that one of the biggest killers in the uk, and indeed the world, lurks. it is called sepsis. so sepsis is when there is a severe infection, and when the body's response to that severe infection leads to organ failure. it can set in without warning, and can affect anyone. this is a lady in her early 50s, she is in day four on the unit, she was very unwell initially. people are admitted to icus for many reasons, but sepsis tends to strike when patients

are at their weakest. we literally collect every heartbeat and every breath. when you first start in intensive care, when i started, it was difficult to see what was important because there was much data. but remember, computers love data, and love spotting patterns in it. they don't get tired, and they never stop. so the doctors and computer scientists teamed up to create something called ai clinician, an algorithm that can detect and even predict early signs of sepsis, and then advise how to stop it. what we have used is a large database with 20,000 patients, and then we tested it in another 80,000 patients. that's more than any doctor could see in their lifetime. the trials were proving successful, but neither the ai or the humans were ready for a new type of sepsis,

one brought on by covid—i9. there are a few differences. one is the intense inflammation they had in the lung made the oxygen levels really low, and so they needed high amounts of oxygen. if they didn't tire of their breathing so quickly as many patients with other pneumonia often struggle with their breathing. we also saw things like a lot of blood clots, which we occasionally see in other sepsis, but were far more frequent in this disease. nearly all the patients coming through to icu now have this new sepsis, and for the ai, this was an illness it had never seen before, nor had any data on. and remember, just because a computer has learned to be good at one task, doesn't mean it can do a different one. the ai clinician was back to square one. but the humans weren't.

their idea was to gather data from icus across the uk to create a database of treatments that worked, so the doctors in the thick of it could learn from each other at speed and save their patients. we build for the first time a picture of how disease was treated in real—time on the databases across the country. and you see all see the diversity of all the different approaches people are trying, what was working and what was not working. now this data will be used to retrain ai clinician and one day it may help to treat sepsis that's caused by covid. so it's about 18 months since we started making this programme and in that time we've all started to use technology in new ways and some of that has included gp consultations over video call, for example.

so i think it probably means that many of us will be happy to accept medical help from machines and computers in the future. but for those computers to be good at theirjobs, they are going to need a lot of data. a lot of our personal medical data. many services like instagram, google and facebook don't charge us any money. instead they get to use our personal data to try and work out how better to target us with ads. for many of us, that seems like a fair exchange. they make our lives easier and they connect us to the world. but now, those same companies which have been collecting and profiting from our personal data are moving into healthcare. amazon, microsoft and google are all trying to grab a seat at the table and dine on our medical data.

and the nhs has something unique to offer. a whole nation's information from birth to death. that's something that might make one feel uncomfortable. i don't think many health professionals sit down with their patients and say do you want your data to be sold to google or deep mined? i don't think those kind of conversations are being had. we're notjust talking about the big household tech names either. plenty of other companies that you never have heard of also want access too. the issue of sale of data to technology companies is an interesting and controversial issue. there are some people who believe it's an asset to the nhs and we should sell it, there are other people who believe that in some way or another that's a wrong thing to do from a moral point of view. but if we get it right — and it is a big if — the benefits could be huge. a unified joined—up store of medical data could really speed up research and lead to new treatments. and ai would be used to look for the patterns hidden within. and that's something that could soon be helping those

who are suffering from long covid. thanks to artificial intelligence, we were able to link patient records with imaging and produce an analysis of 41,000 patients, imaging and clinical associations in about two months. adriana roca is part of a team that has been looking into why long covid only affects some people. what we found was quite important and shocking because it was actually not obesity, per se, but fatty liver disease linked to obesity that was increasing the risk actually up to five times of being severe with covid—19. ajay agade is one of almost a million people suffering from long covid in the uk.

he's a0 and a doctor, and he moved to the uk from india to complete his specialist training. so in the last week of march i was on the weekend duties and i was not feeling well. i went back home immediately. i self—isolated and then i was tested positive. i wasn't thinking that this virus will affect me because i was a fit adult and i thought it would be just 0k, like any other flu virus is for me. there was a moment when i thought i willjust die, and i had a three—year—old, and i would not be able to see her ever in my life. i never knew that this virus will leave me or others with so many problems and disabilities, even after one year. so, to understand his illness more, ajay signed up to a study aiming to help long covid patients — a trial that's using the very same ai that lured me into that big

magnetic tube earlier. this ai had originally taught itself how to spot certain liver characteristics. as well as spotting unusual masses, it had also learned to assess the overall health of the liver. and here's what made of mine. here we've picked up a small area which isjust a cyst, a simple fluid—filled sac. and the ai has picked that up as something unusual, something of interest? yes, but there's no concerning features to it and we wouldn't survey it. and otherwise it's a reassuring scan, with no increased evidence of inflammation or scarring, iron or increased risk of fat. but here's the thing. it turns out that because this ai can assess the overall health of the liver, it can do the same for other organs too. and so we expanded first to organs like the kidneys the pancreas, the spleen and the heart, originally with the aim of looking at type 2 diabetes, but it turned out

that a lot of those organs were being affected by covid as well. if it was a radiographer manually analysing data from, say, six different organs, that could be up to about 24 man— hours of work. this allows us to basically automate some of the most labour—intensive parts of the process. we are trying to identify. which kind of fat is the one that puts you at risk. it's not obesity, as we've thought up until now. - what makes you at higher risk is the visceral fat, _ the fat in the organs. whereas we have found also i people with low bmi with high fat in the liver. ajay is part of the clinical trial which may show that young, fit people with a normal bmi can also get long covid if they have fat in their organs. and not everyone is aware about exactly what long covid is. i think eventually when the acute things wean off, i think more and more people

will come out with long covid. there are many trials in the country trying to understand long covid, and artificial intelligence has become a big tool in the race to find a treatment. but for many, help can't come quick enough. technologies that bring about big change also bring new problems. there is no point in denying that. but right now, ai is changing the world and opening up possibilities for huge medical breakthroughs. i mean, it helped us to develop highly effective vaccines in record time. so i know the pandemic is far from over, but i still feel kind of lucky that it happened now rather than even just a decade ago. and maybe it will mean the nhs is a bit more ready for the next one.

hello there. low pressure brought a very unsettled day on saturday. we saw lots of showers around, some torrential downpours in places, but there was some good spells of sunshine, too. and sunday is a repeat performance — some of the showers will be heavy and thundery once again and slow moving across the north of the uk. that's because in the centre of the low, the winds will be lighter but you'll notice across its southern flank for southern england through wales, quite a squeeze in the isobars there. so, it's going to be breezy. so, we'll start off with maybe some sunshine across eastern areas but the showers will get going. they'll rattle through fairly quickly across southern areas because of the breeze but they'll hang around across the northern half of the country — slow—moving torrential downpours, an area of more persistent rainjust pushing to the far north—east

of scotland and 0rkney there. so the wind is lighter — you'll notice these are the mean wind speeds. further south though, it will be blustering gusts higher than this particularly around coasts and over hills. so in the sunshine, we could see temperatures reach around 21 celsius. 0therwise, like the last few days, most places will be the mid to high teens, and it'll feel particularly cool when the showers come along. now through sunday night, the showers continue across northern areas, turn drier for a while for england and wales but we'll see this feature run across the south bringing some wet weather for south wales and southern england through the night. again, it'll remain fairly breezy in the south, lighter winds in the north. 0vernight temperatures 11—14 celsius. so into monday, then, we're starting the new week off with this area of low pressure but it's going to continue to weaken. fewer isobars on the chart, so the winds will turn a bit lighter. we'll start off quite wet across the south east. that feature clearing away. and you'll see quite a bit of sunshine around and i think general thinking is for monday, there will be fewer showers around but still dotted around, there will be some torrential

downpours in places. top temperatures, again, 20—21 celsius. closer to 17—19 for most. that area of low pressure eventually moves away as we head on into tuesday and a ridge of high pressure builds in across the country from tuesday onwards. this area of low pressure will continue to affect more western parts of the country but i think high pressure will try and dominate for much of the south and the east through the course of the week. so, temperatures will respond. we'll see southerly winds, more sunshine. it will turn much warmer. there will always be the chance of some showers further north and west closer to that area of low pressure.

welcome to bbc news, broadcasting to viewers in north america and around the globe. our top stories: hundreds of wildfires burn out of control in greece. ferries carry people to safety as they flee the infernos. screams for help as residents in the northern suburbs of athens, desperately try to beat back the flames. the fires are still raging out of control. the errors full of the smell of burning smoke and ashes are falling through the air, like snowflakes. gunfire. american b—52 bombers carry out airstrikes in afghanistan in an attempt to stop the taliban seizing more cities. i'm sarah mukerrins live in tokyo on the final day of the olympics, where more gold medals are up for grabs

including in track cycling, basketball and boxing.