Cyberoncologist

Google has developed an interactive microscope for cancer diagnosis

Anton Bugaichuk, Naked Science

Google has invented a system for early diagnosis of cancer in the form of an interactive microscope. The device combines a light microscope and a computer unit, with the help of a neural network analyzes the presence of signs of cancer cell changes. The smart device highlights suspicious areas in real time right in the field of view.

Computer vision technologies, object recognition using neural networks are now used in many fields – from astronomy to observation of laboratory animals. The technique is also suitable for detecting pathological changes in the structure of body tissues. In the simplest cases, a neural network is not needed: there are smartphone applications that can check a suspicious mole for oncology.

Google's smart microscope processes the image in the field of view using a deep convolutional neural network architecture Inception v.3, which already successfully detects melanomas and carcinomas. Neural network training is possible for different types of cancer, when testing the invention, drugs with metastases of breast cancer in lymph nodes (399 samples) and prostate cancer (285 samples) were used.

The figure shows the diagram of the device. The digital camera captures the same field of view as the researcher and transmits the image to the computing unit in real time. The prototype used a frequency of about 10 frames per second – this is enough for smooth movement and magnification of the image.

Smart microscope: diagram on the left, prototype on the right / © Google

The neural network analyzes the video stream, forming a map on which each pixel is assigned the probability of risk of oncological changes. The computer highlights suspicious areas with a line (light green in the figure), you can view the risk map directly.

Identification of the zone of possible cancer changes and a risk map (lymph node, 10X microscope multiplicity) / © Google

The researchers assembled a prototype from a conventional laboratory optical microscope with a multiplicity of 4-40X, having modified it with two modules. A translucent mirror behind the lens passes the light stream further and simultaneously reflects it onto a photo matrix for computer processing. The second module projects the frame onto the original one, the eyepiece shows an already augmented image.

Experimental verification showed the effectiveness of the tool, while recognition was successfully carried out on working preparations of various quality, which is important for clinical use. Therefore, a special microscope is not needed, it is enough to add a block between the lens and the eyepiece to transmit and receive a video signal to models already installed in laboratories.

The inventors point out that such systems can be used not only for the diagnosis of oncological diseases, but also for the analysis of fluorographic and X–ray images, cytology and fluorescence analysis - wherever a person is currently observing laboratory samples directly, and pathology criteria are determined "by eye".

Portal "Eternal youth" http://vechnayamolodost.ru