Mexico. Fujitsu Laboratories Ltd. announced the development of an AI technology that uses deep learning to detect objects, even in cases where only a small amount of data is available.
In recent years, efforts have been made to automate tasks in a variety of fields. In medicine, for example, there has been a desire to use AI to automate tasks, such as detecting objects, including anomalous points, in the analysis of diagnostic images. It is typical to use deep learning in object detection, which involves identifying specific structures in a diagnostic image, but to produce accurate results, tens of thousands of images with correct data are needed. However, since these can only be created by doctors with expert knowledge, it has been difficult to obtain images in such large volumes.
Now, Fujitsu Laboratories has developed a (patent-pending) technology that takes the object's location estimates produced by the object-sensing neural network and converts them into a reconstruction of the original image. Then, by evaluating the difference between the original input image and the reconstructed one, you can create large volumes of correct data where the position of objects has been accurately estimated. This increases the level of accuracy in object detection.
Fujitsu Laboratories has collaborated with the Graduate School of Medicine at Kyoto University and has applied newly developed technology for the detection of bodies called glomeruli (singular glomerulus) in renal biopsy images. The results of an evaluation showed that in an experiment with 50 images with correct data and 450 without correct data, compared to existing training methods that use only the same number of images with correct data, the accuracy of the new technology had more than doubled, under the stipulation of a supervision rate of less than 10%.
Development background
Expectations have been rising for automating tasks using AI in a variety of fields in recent years. Fujitsu Laboratories has been conducting joint research with the Graduate School of Medicine at Kyoto University² and one such initiative has been research to support the diagnosis of kidney disease using AI. In the medical field, there is a diagnostic test that checks the number and condition of structures called glomeruli, which handle blood filtration. The test uses images taken under the microscope of a kidney sample taken in a kidney biopsy. However, it is widely known that there are large variations in the time required for the task of finding glomeruli while enlarging the image, and in how the observer evaluates the state of bodies, even among experts. This has created a demand for automatic and accurate glomeruli number counting and diagnosis.
Issues
To automatically extract the data of possible glomeruli from the images, it is necessary to identify their locations from the images provided and deep learning is known as a method of identifying the locations and types of objects in the images. For this method of training, it is essential to have large volumes of images along with information about the locations and types of objects in them (correct data). However, previously it has been difficult to prepare large volumes of correct data, because it must be created by doctors with specialized knowledge.
Future plans
Through its joint research with Kyoto University's Graduate School of Medicine, Fujitsu will strive to perform the quantitative kidney assessment method by applying the new glomeruli detection technology. This technology is not only applicable to specific applications, such as images of renal biopsies, but also to the detection of objects more broadly, in fields that lack images with the correct data. For example, the company envisions the technology being applied to a wide range of areas beyond healthcare, such as detecting defective products using images of production lines, identifying anomalous locations from diagnostic images using a variety of sensors in infrastructure facilities, and creating bills of materials to be used from architectural blueprints. Fujitsu Laboratories intends to implement this technology during fiscal year 2018 as a learning model building technology compatible with the Zinrai Service Platform, which makes AI technology available through APIs.
