After suffering from a "strange disease" for 3 years and seeking medical treatment to no avail, he was eventually diagnosed withChatGPTSuccessful diagnosis! This is a real experience that happened to a 4-year-old boy. After a certain exercise, he began to experience severe pain in his body. Her mother showed her back and forth17 doctors, from pediatrics, orthopedics and various experts, a series of examinations such as MRI were carried out, but no one really found the cause.
His mother tried to seek help from ChatGPT without high hopes, but the latter directly gave the correct advice based on the description and inspection report.
As soon as the topic came out, it hit the Zhihu hot list, and the Reddit popularity also soared to 2.3k.
Some netizens said that GPT is really exciting this time:
Every doctor's office could be equipped with an AI assistant and compare it to its own diagnosis.
Some netizens took a cue from Med-PaLM, a large auxiliary medical diagnosis model specially trained by Google, and wanted to know its diagnostic results:
For large models, this is almost a perfect test task.
So, what exactly is going on?
What kind of "strange disease" is it?
The protagonist of the story is named Alex, and his mother Courtney has two children.
One day in 2020, Alex's nanny told Courtney that Alex had to take painkillers every day, otherwise he would collapse from the pain.
Then, Alex started grinding his teeth again. His parents linked the two things together and thought it might be caused by pain caused by tooth replacement or tooth decay.
So my mother took Alex to seedentisit, Alexup to three yearsThe journey of seeking medical treatment officially began.
The dentist did not find any problems after the examination, but suggested that since Alex was grinding his teeth, he recommended that they see a professional to treat airway obstruction.Orthodontist.
The orthodontist discovered that Alex's upper jaw was too small, making it difficult to breathe, so he placed an expander on Alex. This treatment method did have an effect, and the mother once thought that Alex's disease would be cured soon.
It's reasonable, but reality is often illogical.
Her mother soon discovered that Alex, who was only four years old, had suddenly lost weight.
This time, my mother asked for help from apediatrician.
Doctors thought Alex might have been affected by the coronavirus, but his mother was not satisfied with this explanation. However, my mother still took Alex for a review in early 2021.
The doctor told his mother that Alex had "grown a little taller", but found that Alex's left and right feet were somewhat unbalanced and suggested that they choose physical therapy.
This time, her mother believed the doctor's words, but before physical therapy began, Alex developed headache symptoms again, which became increasingly severe.
The matter of physical therapy had to be put on hold for now, and my mother took Alex to see him first.Neurology (internal medicine)Doctor, the conclusion is that Alex suffers from migraines.
While fighting headaches, Alex was also troubled by symptoms of exhaustion, so he was taken toOtolaryngologyCheck with your doctor to see if there are any sinus problems that are affecting your sleep.
After experiencing these twists and turns, Alex finally began to receive physical therapy.physical therapistIt is believed that Alex may have a congenital disease called Chiari malformation.
This congenital disorder causes abnormalities in the brain where the skull meets the spine.
The mother began to research this and took Alex to see a new pediatrician.Pediatric Internal Medicinedoctor,Adult Internal Medicinedoctor andMusculoskeletalDoctor etc.
In the end, Alex saw as many as 17 doctors. It can be said that he visited almost every department imaginable, and was even sent to a hospital.Emergency, but still couldn’t find out why.
Until ChatGPT caused a 180-degree reversal of the entire incident.
With the mentality of giving it a try, my mother registered a ChatGPT account.
She entered Alex's symptoms and notes from the MRI report, including one detail that Alex was unable to sit cross-legged.
ChatGPT gives the diagnosis——tethered cord syndrome(TCS).
Of course, Courtney didn't believe it directly. After getting the answer, she first found a communication group for parents of children with children on Facebook.
After reading the discussion, my mother felt that these symptoms were very similar to Alex's.
This discovery rekindled the almost extinguished fire of hope. Afterwards, my mother recalled that she sat in front of the computer all night and experienced everything.
With this conclusion and Alex's MRI report, foundneurosurgerydoctor.
This time we finally found the right person. The doctor took a look at the MRI and came up with the same conclusion as ChatGPT, and pointed out the specific location of the tether.
Things went more smoothly after that. Alex underwent surgery and is currently recovering.
So why didn’t Alex get the final diagnosis until he saw the 18th doctor?
first, and it also has a certain relationship with Alex himself.
People with TCS often develop clefts in their backs, but Alex did not have a cleft, a condition called occult tethered cord syndrome (OTCS).
Although TCS is a rare disease, its incidence in newborns is not low, about 0.005~0.025%, which is higher than the incidence of leukemia.
Chen Yingge, Mi Yang. A case of multiple fetal development abnormalities during pregnancy [J]. Progress in Clinical Medicine, 2023, 13(2)
But OTCS is relatively rare—so rare that the incidence rate has not been calculated at all.
But after all, at the end of the story, the surgeon made a quick judgment after seeing the MRI images.
so, it could not be diagnosed before, perhaps because of "finding the wrong doctor": among the 17 doctors, none of them were engaged in surgery.
Of course, this is normal. After all, they are all specialists (corresponding to general practitioners) who are good at their respective fields of expertise. It is inevitable that they will not have a comprehensive understanding of knowledge outside their majors.
But this also exposed a problem. When encountering unexplained problems, these doctors did not considermultidisciplinary consultation, it is also unknown whether Alex’s medical history has been fully inquired about.
In the words of her mother, Courtney, no one was willing to solve "the bigger problem (beyond their own discipline)" and no one would give any clues about the diagnosis.
ChatGPT's knowledge base is at least much richer in breadth than professionals in subdivided fields. It takes Alex's situation into consideration more comprehensively and finally gives the correct conclusion.
So is ChatGPT's successful diagnosis this time a mistake, or does it actually have diagnostic capabilities?
Can AI be used for diagnosis?
In fact, this is not the first time someone has used ChatGPT or GPT-4 as a disease diagnosis tool.
For example, soon after GPT-4 came out, someone used it to successfully diagnoseown dogA kind of case, this experience once exploded on the Internet.
He told GPT-4 the dog’s symptoms from the first onset, the treatment process, and each blood test report:
On the 20th, she had a high fever of 41.5 degrees Celsius. The doctor diagnosed her with canine babesiosis based on the blood test results (blood test results attached). She received antibiotic treatment for the next three days. On the 24th, she received antibiotic treatment, but her gums turned pale (new blood test results attached).
GPT-4 quickly gave the detection results and indicated in the conversation that it may be caused by the following two reasons:
1. Hemolysis: Destruction of red blood cells due to various reasons, such as immune-mediated hemolytic anemia (IMHA), toxins, or other infections other than babesiosis.
2. Blood loss: Internal or external bleeding, which can be caused by trauma, surgery, or gastrointestinal problems (such as ulcers or parasites).
The doctor's final diagnosis showed that the dog was indeed suffering from immune-mediated hemolytic anemia (IMHA), and the dog was saved after the right medication was given.
In addition, some netizens reported that they were hacked by ChatGPT (GPT-4).saved a lifeexperience.
He felt sore all over after going to the gym. After consulting GPT-4 about his condition, he got the answer of "rhabdomyolysis" and immediately went to the hospital and died as a result.
However, some academic studies have mentioned that neither ChatGPT nor GPT-4 is an AI doctor that can be completely relied on.
For exampleHarvard University HospitalA study from Brigham and Women's Hospital (BWH) published in JAMA showed that when giving cancer treatment recommendations, ChatGPT was completely correct only 62% of the time.
In other cases, 34% of the suggestions contained at least one or more answers that were inconsistent with the correct diagnosis, and 2% of the cases gave an implausible diagnosis.
In this regard, the study believes that the diagnosis cannot be completely left to ChatGPT or GPT-4. After all, they still cannot compare with professional doctors in the diagnosis process.
(However, some netizens pointed out that the reason why ChatGPT failed to diagnose successfully may also be related to the training data, and treatment information after 2021 is not included)
In this regard, Andrew Beam, assistant professor of epidemiology at Harvard University, believes that the effects of using ChatGPT and GPT-4 should be viewed from two sides:
On the one hand, they are better to use than some ordinary diagnostic software or Google search engines, especially the GPT-4 version.
But on the other hand, they are not yet likely to replace clinicians with substantial expertise. After all, for AI, it is indeed possible for them to fabricate information when they cannot find the answer, and infer wrong results based on "hallucinations".
Jesse M. Ehrenfeld, president of the American Medical Association (AMA), said that even if AI can diagnose the results, the ultimate responsibility lies with the doctor himself.
To sum up the above point of view, everyone can use AI to assist in diagnosing the disease, which is easier to use than search engines, but in the end you still have to go to the hospital and find a doctor for diagnosis.
So, if you plan to use a large model for "inquiry consultation", which large model is best to use?
Some netizens used themselves as a case to test whether various large language models have diagnostic capabilities, and finally thought thatGPT-4Comparatively capable of:
I consulted several doctors about the cause of my chronic cough, but finally learned on a YouTube channel that I had LPR (Latent Laryngopharyngeal Reflux).
I tested the large model using my own cases,GPT-4was the only one successfully diagnosed. Although Claude2 had a close answer, he was not able to diagnose it completely independently.
Have you ever tried using AI to help diagnose your condition? How does it feel?