Artificial Intelligence Can More Accurately Detect Lung Cancer and Heart Diseases Than Doctors
Over the past couple of years, the use of AI in detecting various health ailments, have proven useful in the health sector. Doctors no longer find it difficult in identifying patients diagnosed with heart diseases and lung cancer as these AI systems have this technique pretty easy.
Despite the fact that cardiologists are very good at their jobs, they are also capable of making mistakes. In order to determine if a patient’s heart is properly functioning or not, a cardiologist will have to analyze the timing of their heartbeats in scans. Analysis has it that approximately 80% of these scans are correctly carried out with about 20 % indicating some errors and enhancement. This lead to the development of Ultromics, an AI diagnosis system that has been proven to be more accurate compared to doctors. This system was designed and developed at the John Radcliffe Hospital in Oxford, England.
Ultromics was tested using the heart scans of over a thousand patients. These patients were treated by the company’s chief medical officer, Paul Leeson. The system was also able to detect if the patient has suffered any form of heart problems. In an interview with BBC NEWS, lesson beamed that the system has been tested in multiple clinical trials and has outperformed human cardiologist in all trials.
However, Optellum is working hard to commercialize an AI system with the ability to detect patients diagnosed with lung cancer by analyzing clumps of cells found in scans. Timor Kadir, the company’s chief science and technology officer in an interview with BBC NEWS, stated that the system has been widely tested in various trials and could be used in testing over four thousand lung cancer patients yearly.
PRESERVING LIVES AND MONEY
Asides the fact that AI diagnosis systems could be used to save lives by providing earlier diagnosis of heart problems as well as cancer, they could also be more cost-effective by preventing one from spending lots of money in hiring more doctors, nurses, and other hospital staff. “Optellum could cut costs by about £10bn ($13.5 billion) if both the United Kingdom and the United States decides to make use of it,” said Kadir in an interview with BBC NEWS. Similarly, the United Kingdom healthcare tsar Sir John Bell told BBC NEWS that there is every possibility that AI would have a tremendous positive impact on the National Health Service (NHS). He further said that about £2.2bn ($2.97 billion) is being spent on pathology services annually and AI may be able to reduce this cost by 50 percent if used.
But unfortunately, some medial personals are of the opinion that these AI systems might be able to replace doctors in the nearest future. However, given the numerous roles of doctors in the healthcare sectors, these systems are more likely to play a supporting role to doctors and will assist human workers to effectively and efficiently carry out their tasks.
Using AI to Detect Lung Cancer from CT Scans
Medical scientists have beamed that to detect, diagnose, and treat a patient with cancer successfully, access to the right tools is essential. For instance, in England, lung cancer which happens to be one of the significance causes of death with an annual 1.6 million deaths attributed to it, is most times detected at a late stage living majority of these patients untreatable.
To diagnose lung cancer, physicians depend on the segmentation of lesions on the lungs by making use of a combination of PET and CT scans. These are primarily used to determine the functional attributes of a lesion in addition to its anatomical characteristics and structure.
Poland-based Future Processing which happens to be one of the numerous members of NVIDIA’s inception program is working really hard to ensure it simplifies the use of these tools, thereby making the diagnosis process more economical, accurate, reliable, and accessible.
Furthermore, its medical imaging solutions business sector has collaborated with the medical imaging experts, research institutes and clinics all over the world to design and develop software that will enhance and adequately analyze images.
Dynamic contrast enhanced imaging, and the analysis of computed technology (CT), happens to be their significant areas of concentrations. The company’s research in this area is likely to lead to the increased utilization of CT scans in the discovery and examining of lung cancer.
A Picture Expresses A Thousand Words
Future Processing is working so hard to determine the best solution to the get rid of the need for the unification of PET and CT scans. This will enable doctors to make diagnosis and analysis based on CT scans.
By making use of the convolutional neural networks, they have been able to demonstrate that diagnoses from CT scans alone can be made efficient and accurate.
According to Dr. Jakub Nalepa, a senior research scientist at Future Processing, “Before the segmentation of active lesions there is the need for the co-registration of PET and CT sequences within a time-absorbing procedure.” He further stated that they are have been able to present a research analysis using CT scans, where they were able to demonstrate segmentation of a single image within a couple of minutes.
NVIDIA Tesla GPU accelerators are responsible for powering the acceleration in speed segments and is capable of making a tremendous difference for both doctors and patients. Furthermore, radiologists can save a tremendous amount of time and measure lesion progress by automatically segmenting the lesions.
It would be a huge relief for medical sites as they will be able to take care of their patients directly using only CT scanners and without the interception of PET scanners. This is pretty more economical for medical personals, as a CT scan costs between the sums of $1,200 to $3,200, while a PET scan which costs $3,000 to $6,000. Asides that, it also offers a better experience for patients, who will only undergo one scan.
Nalepa and his team of researchers have proven their approach to be responsible for the reduction of false positives, from 90.14 percent to 6.6 percent. They hope to develop a lasting solution to the lung cancer in the nearest future.
New Algorithms are Able to Diagnose Disease as Accurately as Expert Physicians
Sebastian Thrun and colleagues at Stanford University earlier this year demonstrated the capability of ‘deep learning” algorithm; this deep learning algorithm was capable of diagnosing potentially cancerous skin lesions as efficient as a board-certified dermatologist.
Professionals have beamed medical images, such like x-rays, MRIs, and photography; to being an almost a perfect match for strengthening deep-learning software, which in the pasts has led to the breakthrough in recognizing faces and objects.
A good number of companies are already in pursuit. Alphabet life science arm, merged with Nikon last December to design an algorithm for detecting the causes of blindness in diabetics. However, the radiology field has been titled the “Silicon Valley of medicine” owing to the numerous numbers of detailed images generated by it.
Though the prediction made by the Thrun team were highly accurate, it wasn’t sure exactly which of the features of a mole, the deep-learning program uses to classify it as cancerous or benign. The outcome is the medical version of what is termed deep learning “black box” issues.
But unlike the traditional vision software, where rules are defined by a programmer, for instance a stop sign, it has eight sides and in deep learning, the algorithm searches the rules itself, but often do not leave an audit trail to explain its decision.
A scholar from the University of Michigan, Nicholson Price, who focused on health law, says “In the case of black-box medicine, doctors do not know what is happening because nobody does its heritable opaque.”
Yet he says that, that may not stand as a severe hindrance in health care; it relates deep learning to drugs whose benefit comes from unknown means. Lithium is an example, the biochemical and mechanical effect of mood is yet to be elucidated, but the drugs are still approved for the treatment of bipolar disorder. The mechanisms behind the most widely used medicine aspirin weren’t understood until after 70-years.
Price has stated several times that the black-box issues won't be a problem with the U.S. Food and Drug Administration; and in the aspect of approving new drugs and as well regulating software if its purpose is to prevent or treat disease.
The FDA says in a statement that for over twenty years it has approved numbers of image analysis application which relies on a variety of patterns recognition, in machine learning and as well computer vision techniques. More software powered by deep learning has been confirmed by the agency and noted that companies are allowed to keep details of their algorithm confidential.
The FDA has already approved and given the freedom to at least one deep-learning algorithm. In January the FDA open for the sale of software developed by a privately held medical imaging company resided in San Francisco. The algorithm, “Deep Ventricle” analyzes the MRI images if the interior contour of the hearts chambers and as well calculated the volume of blood a patients heart could contain or pump. The calculation was completed in less than thirty seconds, and Arteries says whereas conventional methods typically would take an hour.
FDA expected Arteries to do further testing to make sure his algorithm result was on par with those generated by physicians. John Axerio-Cities said he needed to prove the statistics that Arteries algorithm followed whatever its intended use is, or what the marketing claims to say its doing.
The team led by Thrun a former vice president of Google who also works on driverless cars, n training their software fed it 129,405 images of skin conditions evaluated by experts. These covered the number of 2,032 various diseases and included 1,942 images of confirmed skin cancers.
The software eventually was capable of outperforming 21 dermatologists in identifying mole which was potentially cancerous.
Robert Novoa, a Stanford dermatologist and an author of the study, has said that when most dermatologist sees the potential of this technology, they will embrace it. Robert Novoa and some other team members refuse to say if its in their plan to commercialize the software
Allan Halpern a Memorial Sloan Kettering dermatologist and president of the international society for digital imaging of skin, he has debunked the worries that doctors would soon be out of a job, “I think the threat is the opposite,” he said. The algorithm could dramatically increase the demand for dermatologist services in the society.
Halpern as well let us know that it’s because a positive screening test still needs a biopsy, and deep-learning software could find a role in primary care offices, but if it be that it was made available as a population-wide screening test or through online app, their won't be enough dermatologists to attend or follow up on the leads.
Axerio-Cities says companies will be tempted to offer deep-learning tools directly to their customers, for example people may scan their moles to see if they actually need to visit a doctor, like some non-Al cell phone apps, like MOLE MAPPER, can be used to track suspicious moles in the body, an as well record any changes over time.
However, Halpern says that customers may likely be not ready to deal with a diagnostic system that might notify them of a mole, or telling them a mole has 5 percent chance, or 50 percent chance of being cancer.
Google's Deep Learning Algorithm Is Already Better at Diagnosing Diabetic Retinopathy than Many Physicians
One of the major breakthroughs in Health Care and Artificial Intelligence world happened recently when Google successfully trained a Neural Network with 120,000 retinal images to detect diabetic retinopathy.
Generally speaking, Diabetic Retinopathy (DR) is one of the most common causes of increasing cases of blindness in the world. Studies reveal that high blood sugar level can damage the retinal blood vessels that can further cause irreversible blindness. Hence, early detection of this disease was need of the hour. But the sad fact is that physicians were not able to recognize the symptoms at an early stage. Hence, Google decided to use Machine Learning algorithms to solve this trouble. The major target was to identify disease symptoms before they cross the critical boundaries so that an early diagnosis can be carried out.
Google trained a convolution neural network for image classification and to achieve the top accuracy they used a huge set of sample data. The training dataset contained 128175 retinal images collected from EyePACS in the US and three different eye hospitals of India. These images were graded by 3-7 licensed ophthalmologists. After training the Deep Learning Neural Network, two validation tests were performed. The first test was performed with 9963 images collected from EyePACS and the second dataset of 1748 images was taken from Messidor-2. The performance of this machine learning algorithm was later compared with the ophthalmologists and the results were observed to be quite efficient.
The scientists at Google targeted two parameters for this application, they were sensitivity and specificity. The ultimate results after training provide around 96-97% sensitivity and 93% specificity.These results show that artificial intelligence suggests a way to utilize medical solutions in the much creative way. Deep learning is a bright concept for the world of technologies as it can ensure early detection of diseases so that vision of the person can be saved on time. This revolutionary research shows that AI can perform much better as compared to most eye pathologists. Moreover, most corners of the world don’t even have experienced medical health professionals to treat patients, in such situation; the automatic detection is a great solution for people suffering from various diseases. The great news is that ability of Deep Learning is not just limited to detect DR rather it can show impactful results for many other medical health issues as well.
Clinical Trial Involving 10,000 Chest X-rays that were Successfully Analyzed by AI