Source: ATYUN AI platform
Image analysis study supported by the machine can accurately predict chronic kidney disease in patients with renal survival time.
Recently, a study "Kidney International report" published showed that kidney biopsy machine learning and image analysis helps to help predict the time of a fully functioning kidneys in patients with chronic renal impairment.
The researchers used the depth of learning and neural networks (a kind of imitation machine learning decision-making model of the human brain) found in the calculation of renal function decline, a new series of convolutional neural network (CNN) algorithm pathology scoring system than the traditional estimate more precise.
Boston University research team, explains: "In the biopsy sample, usually by chronic kidney damage in renal biopsy samples, the number of fibrosis and tubular atrophy get to the semi-quantitative assessment."
"Although pathologists trained eye can determine the severity of the disease, and are able to detect the nuances of the amazing accuracy of histopathology, but this expertise is not able to get everywhere, especially in the globally."
According to the US National Institutes of Health show that in the United States, about 14 percent of people with chronic kidney disease. This usually rare symptoms until it raises the importance of regular monitoring, and accurately determine the progress of the disease.
"In addition, the urgent need for the severity of the pathology of the disease is normalized in order to establish the effect of treatment in clinical trials, can be used in daily practice in the treatment of patients with less serious illness." Boston University research team added.
Artificial intelligence and image analysis to help pathologists to complete these tasks technology. Artificial intelligence can recognize multi-GB pixels in an image pattern, provides a detailed analysis of large amounts of data level is incredible human clinical doctors can not match.
The early results of machine learning and image data combining exciting. Some pilots have demonstrated that artificial intelligence tools and human pathologists almost as accurate, while significantly reducing the time needed to analyze large amounts of data.
In order to achieve the same positive results, the research team at Boston University for the data of 171 patients between 2009 and 2016 Boston Medical Center for treatment were investigated.
These patients represent typical of chronic kidney disease patients, mean age was 52 years, while suffering from complicated by chronic hypertension, diabetes and so on. Of which nearly half of the population is African American.
The researchers used Google's image recognition Inception V3 architecture, millions of images are pre-trained to support changes available biopsy slice identification.
The algorithm is trained to identify possible kidney survival rates of 1 year, 3 years and 5 years. Since this study using retrospective data, the team is able to predict the results of the algorithm and the actual match.
The results showed that the survival rate in predicting kidney three goals periods, CNN model is much better than the pathologist estimated scoring system. The algorithm can more accurately identify an individual state kidney disease.
Source: Report of the International Kidney
"An important advantage of this study is a machine learning technique is applied to conventional biopsy specimens trichrome-stained tissue image without the need for any special handling or digital scanning operation," the research team said, "This allows us to directly results of comparative analysis of machine learning and clinical pathology report from the same specimen. "
The use of learning technology also helps to create depth than the pathologist estimated scores more complex assessment framework, scores depends on the degree of fibrosis in particular present in the sample.
The study said: "The use convolution, activation and centralized (or resampling) and other operators, training CNN model many times as necessary to perform these operations in a systematic way, to pixel-level information into the advanced features of the input image. "
CNN model training using a single value (fibrosis score) as the input contrast features and output class corresponding to the model training pathologist completed. This aspect highlights the value of using computer algorithms (such as CNN) to capture the pixel-level information from the entire image, and its associated associated with interesting results, rather than fibrosis score itself.
The researchers pointed out that despite CNN's model is better than a simple score, but the pathologist estimated fibrosis score is still monitoring the progression of chronic kidney disease is a very valuable and accurate way.
The study said: "Obviously there are limitations of machine learning algorithms, and provides incremental value, not replace the human factor." "We recognize that clinical manifestations and diagnosis of nephrologist is based on the relevant factors, and not in isolation of the lesion and pathological visual inspection. "
"However, the accuracy of computer use experienced renal pathologist for histological images ability to categorize, may affect the practice of the kidneys, especially in resource-limited environment."
With machine learning algorithms become more complex, suppliers may soon be able to find their deeper integration into clinical decision support tool to guide treatment decisions.
The study concluded: "This fast, scalable method may be expanded in the form of software in the care of the time, and has the potential for substantial clinical impact, including increased clinical decision kidney scientists."
"In different clinical practice and the image data set to further validate the model, for all of the distribution and extent of lesions encountered in a typical pathology services, to verify that this technology is necessary."
This switched ATYUN artificial intelligence media platforms, the original link: medical breakthrough! Can accurately predict survival time by kidney machine learning techniques
more recommendations
Computer vision can confirm the intent of the pedestrian crossing, traffic efficiency
NVIDIA launched edge AI computing platform EGX, real-time optimization of AI services
Every Dajiang Innovation penetration action camera to capture amazing detail of 4K adventure
