The "amateur" Programmer Used 50 1080ti To Fight Cancer

take 26 minutes to read
Home News Main article

"That's what programmers should look like!" "The landlord is really strong. Technology saves the world" "compared with him, I feel that the code I wrote is meaningless..." these netizens' comments came from a post in V2EX forum, the gathering place of programmers. When you see these messages for the first time, you may think they are a little secondary, but for those who have been saved by posts and families, such comments are worthy of the name.

Because the purpose of the post is to detect breast cancer.

In 2018, a landlord named coolwulf opened a post saying he had built a website. As long as users upload their own X-ray images, AI can quickly diagnose the condition of breast cancer for themselves. The accuracy of tumor recognition has reached 90%.

In short, let AI help you "watch the film". The accuracy is almost comparable to that of professional doctors, and it is completely free.

As we all know, although the cure rate of breast cancer is high, because the early symptoms are not obvious, it is easy to miss the best time to cure. It is often found that it is in the late stage.

However, a reliable tumor detection AI can let a large number of patients who cannot see a doctor in time know their condition in advance. Even if the final diagnosis needs a doctor, it is precious in many areas with tight medical resources.

Meanwhile, the incidence rate of breast cancer is also the highest among all cancers

Soon, coolwulf received hundreds of rare replies to this post. In the comment area, someone is waiting for the doctor's test results, anxious;

There are also netizens whose families suffer from breast cancer, which is full of unknown and fear. And coolwulf's project gives them hope.

What follows is, of course, everyone's curiosity about the project and coolwulf himself. Is AI reliable in detecting breast cancer? Where does the huge amount of clinical data and hardware computing power come from? More importantly, who is the super God who is willing to open it for free?

Coolwulf didn't answer many questions one by one. It even happened very soon. I brushed my clothes and hid my skills and fame. It rarely appeared again. Who knows, in 2022, his hero returned with a heavier "brain cancer project", and the mystery remains.

In order to dispel the fog of coolwulf, the bad critic contacted him as far away as Central America. After several rounds of interviews, today, let's listen to the story of coolwulf - coolwulf, formerly known as Jiang Hao.

As a student, he studied for undergraduate and doctoral degrees in the Department of physics of Nanjing University and the Department of nuclear engineering and radiation science of the University of Michigan. He commented on his career briefly: "although his main business is imaging medicine, he is also a programmer working on open source projects in his spare time.".

Coolwulf Jiang Hao (first from the right) ▼

If you list his own resume, it will even make people wonder if he has any super abilities. Coolwulf's parents were not medical practitioners, but they developed his interest in programming from an early age. Therefore, coolwulf likes to write some code in his spare time. In those days when there was no GitHub, he often put his projects in the old programmer community of SourceForge.

He joined an open source project called Mozilla Foundation around 2001. At that time, this group had two initial projects, one of which was k-meleon (a browser that was popular in China in the early years) for which he contributed code.

Another project, code named Pheonix, is the predecessor of the familiar Firefox browser. Also because of this, he was interviewed by the media more than ten years ago.

Coolwulf also wrote a website in 2009, which can help everyone book hotels at low prices. Many foreign students in North America should have used it. And these are just the result of his spare time and interest.

After completing his studies in imaging medicine, he successively served as the R & D director of Brooke and Siemens to guide the product development in the imaging field. A few years later, he and Lu Weiguo, now a lifelong professor at the Southwest Medical Center in Texas, founded two software companies for the radiotherapy market and began to engage in product research and development of cancer radiotherapy and artificial intelligence technology.

In short, throughout the majority of coolwulf's life, it should be the "learning God" in our mouth. We haven't run away. PS: my brother not only blossomed in his studies and sidelines, but also was the main point guard of the basketball team when he was in Nanjing University.

Brooke photon III ▼ led by coolwulf

Maybe if he continues to develop like this, he will become a scientific research entrepreneur far away from us. But the following is not only the turning point of coolwulf's life, but also the starting point to shorten his distance from thousands of families and life.

His 34 year old alumnus of Nanjing University, who missed the best treatment period for breast cancer, unfortunately died, leaving only a 4-year-old son. Coolwulf regretted after witnessing life and death and the families destroyed by illness. At the same time, he also learned that many breast cancer patients often lack detection channels, which is easy to delay the treatment.

Old photos of alumni family ▼

So coolwulf came up with the idea of using AI to detect X-rays. He also happens to have corresponding professional experience. However, it is not easy to be an AI that can accurately detect tumors.

Coolwulf first downloaded ddsm and MIAs data sets on the official website of the University of Florida. However, due to the long history of the data and the fact that the image data are still reproduced on film, a special program is written to turn all the information into usable form.

Then, in order to obtain the breast cancer data set, a non-public resource of the University of Barcelona, he specially wrote an email to ask for permission. During this period, we should also continue to read a lot of literature and write the corresponding model code.

Request email sent by coolwulf at that time ▼

However, this is not enough. To formally train this model, it often requires very high hardware computing power. Therefore, at his own expense, he built a GPU computing set composed of 50 NVIDIA GTX 1080 Ti locally.

It is true that these 50 graphics cards are hard won. Because in the process of buying, we have to fight with the scalpers. Coolwulf found friends, used many credit cards and kept refreshing on the website. So far, we have finally completed the preliminary preparation of the website.

Yes, in addition to games and mining,

There are more uses for graphics cards ▼

This free AI breast cancer detection website spent about three months of spare time before and after coolwulf, sleeping directly in the office when busy, and finally launched in 2018.

He and poor review Jun said that because the background does not save data, it is not clear how many people have used it. But during that time, he received many thank-you emails from patients, many from Chinese. Moreover, users really use the website to check out tumors, especially for people in areas with limited medical resources, which is equivalent to robbing time from death.

The first guy got the picture wrong,

Tumor was found after retest ▼

A few years ago, the relevant technology had not been popularized, so the coolwulf project was an enlightenment. The website also received extensive attention in the industry. During this period, there were many medical institutions at home and abroad, such as surgery of Fudan University, who thanked him by email and were willing to provide financial and technical support.

After all, the whole thing coolwulf is at one's own expense, which is not a small amount.

Email from Fudan University Cancer Center ▼

As for why not use the website commercially and charge some money, you also asked this question.

Coolwulf's reply was indifferent but noncommittal: "cancer patients and their families have suffered too much. I believe everyone wants to help them, and I just have the ability." in this way, he thanked many people for their financial assistance and took care of everything on his own.

In addition to the website, there was a desktop version of the detection software ▼

Time to 2021, for coolwulf, also ushered in the second key turning point in life. His colleague's cousin suffered from brain tumor and was not optimistic. He was treated with "whole brain radiotherapy". Unfortunately, a few months after whole brain radiotherapy, the patient's brain tumor relapsed again, but there was no treatment at this time and had to wait for death.

Whole brain radiotherapy is a kind of therapy that can disinfection and sterilization patients' brain tumors in a wide range through radiation. It can not only eliminate cancer cells, but also cause damage to normal brain cells, so as to reduce the occurrence of lesions.

In less rigorous terms, whole brain radiotherapy is more like an "indiscriminate attack". Therefore, considering the tolerance of the brain, whole brain radiotherapy can only be done once.

After this, coolwulf completely changed his view and decided to break through an industry problem - no longer let AI stay in the detection stage, but really put it into practical treatment.

You know, whole brain radiotherapy is the most common treatment for brain tumors. In the United States, 200000 people receive whole brain radiotherapy every year. So, do patients with multiple tumor brain cancer have to take risks and choose whole brain radiotherapy?

In fact, it's not necessarily true, because there is another treatment - stereotactic radiotherapy. Compared with the large-scale irradiation of whole brain radiotherapy, the radiation of stereotactic radiotherapy is more concentrated, which can accurately remove the diseased tissue without damaging the normal tissue.

Like our familiar gamma knife, it is a kind of stereotactic radiotherapy. This therapy has low side effects and less harm to patients, and can be used many times.

In the academic circles, it is also generally agreed that stereotactic radiotherapy can provide patients with better quality of life and more significant curative effect. The only problem is that stereotactic radiotherapy will make medical resources more scarce.

Because once this scheme is adopted, oncologists must accurately outline and mark each tumor; Physicists also have to make accurate treatment plans for each tumor. It will take a lot of time to treat a patient.

Therefore, when patients have too many tumors, doctors almost tend to choose whole brain radiotherapy rather than stereotactic radiotherapy.

However, AI imaging technology may be used to share the workload of doctors. Therefore, in order to make more brain cancer patients can use stereotactic radiotherapy, coolwulf once again invested in research and development.

However, the problems faced this time are obviously more challenging, and he can no longer do it alone. So he found Southwest Medical Center and Stanford University to seek cooperation.

With the help and efforts of many people, after many years, the following three AI models were finally developed not long ago: the model of automatically outlining / labeling brain metastases; Based on SVM radiomics, false positive models can be quickly reduced; Based on the optimized radiation dose map, the model can quickly segment multiple lesions into different treatment courses.

The three models complement each other and correspond to the doctor's workflow, which can greatly reduce the workload when using stereotactic radiotherapy.

This project has been published at the AAPM spring clinical meeting in 2022, and has once again been widely recognized in the industry.

Coolwulf and collaborators are also speeding up their efforts to make the whole stereotactic radiotherapy community available. In the interview, coolwulf mentioned many times that today's achievement is by no means due to him alone.

He hoped that the bad critic could publish the list of partners, because everyone here is a hero fighting cancer silently.

Cancer mortality in recent years has fallen by 30% compared with 30 years ago. According to this development, perhaps one day in the future, cancer will no longer be a terminal disease.

But this is not a natural thing, but there are countless people like coolwulf them, carrying lights in the abyss. At the end of the article, let's borrow the comment of a foreign netizen under reddit as the end:

"Not all heroes wear cloaks.".

NVIDIA Was Fined $5.5 Million For Misleading Investors In The Cryptocurrency Market
« Prev 05-07
Musk's Seven Candidates For CEO After Acquiring Twitter: Founder Dorsey Is Not Among Them
Next » 05-07