Artificial Intelligence in Smartphones: 5 Technologies for Everyday Use

Contents:

• Artificial intelligence recognizes cough
• Innovations in mole diagnostics
• Detecting eye diseases using technology

Artificial Intelligence Recognizes Cough

The AI ​​Resp app, created as part of the SberMedII initiative, represents a significant step forward in COVID-19 diagnostics. It uses cough and voice analysis to detect coronavirus infection with up to 82% accuracy. This innovative technology allows for quick and effective health assessments, which is especially important during the pandemic. AI Resp provides access to diagnostic data, making the COVID-19 screening process more convenient and accessible for users.

AI Resp is available for free download on Google Play and the App Store, making it accessible to a wide audience of users.

The creation of an app capable of diagnosing COVID-19 became possible thanks to research conducted by specialists at the Massachusetts Institute of Technology (MIT) in September 2020. In their papers, they presented an algorithm that analyzes audio recordings of patients' coughs and voices to identify signs of the disease. This discovery marked an important step in the use of technology for early diagnosis and health monitoring. The development of such an app could significantly improve the process of detecting COVID-19 and contribute to a more rapid response to the pandemic.

MIT has extensive experience using artificial intelligence in the field of medicine. In particular, researchers successfully adapted a neural network originally developed for diagnosing Alzheimer's disease to detect cases of COVID-19. These innovations highlight the potential of artificial intelligence in improving medical diagnosis and treatment of various diseases.

The specialists collected an extensive dataset to train the new model, including five thousand cough and voice recordings of both COVID-19 patients and healthy individuals. The creation of the OpenSigma platform allowed anyone to contribute by providing cough recordings. This collaboration has significantly expanded the database and improved the accuracy of model development, which can help in the diagnosis and monitoring of COVID-19.

The neural network developed at MIT has demonstrated high accuracy in classification COVID-19, achieving a 98.5% success rate. This is particularly important because the system is capable of detecting the virus even in its early stages, when patients do not yet have obvious symptoms. Despite these impressive results, the algorithm is still undergoing clinical trials at medical facilities specializing in COVID-19 treatment. This approach has the potential to significantly improve diagnosis and early detection of the disease, which in turn will aid in the fight against the pandemic.

At the beginning of the pandemic, Sber's Artificial Intelligence Lab developed a neural network capable of analyzing the extent of lung damage using computed tomography (CT) scans. This innovative technology enables rapid and accurate assessment of patients' condition, which is critical during a medical crisis. The neural network was trained on a large volume of data, ensuring its high effectiveness in diagnosing lung diseases, including COVID-19. The development of this system highlights the importance of artificial intelligence in medicine and its role in improving the quality of diagnosis and treatment.

Sber has developed an algorithm that converts audio files into visual graphs—spectrograms. These spectrograms display air vibrations, allowing for visual analysis of audio data. This approach opens up new possibilities in sound processing and information visualization, which can be useful in various fields, from music to science.

Three thousand recordings of coughs and voices of COVID-19 patients were used to train the neural network. The training set also included recordings of healthy individuals and various background noises. This enabled the neural network to effectively distinguish between coughing, breathing, and voice sounds associated with COVID-19. This approach improves diagnostic accuracy and facilitates earlier detection of the disease, which is especially important during a pandemic. Using various sources of audio data helps improve machine learning algorithms and their adaptation to real-world conditions.

The AI ​​Resp app improves the accuracy of COVID-19 diagnostics by prompting the user to report symptoms such as fever and headache. This allows for a more effective assessment of the risk of illness and recommendations for further action.

Innovations in Mole Diagnostics

The ProRodinki mobile app offers an innovative solution for the early detection of malignant tumors that can masquerade as nevi. Using modern artificial intelligence technologies, the app effectively aids in the diagnosis of skin diseases such as basal cell carcinoma and melanoma. This app not only makes it easier to monitor your skin condition, but also helps raise awareness about the risks associated with skin lesions.

The app is available to all users and can be downloaded free of charge on Android and iOS devices. You can download it for Android and for iOS.

Dark formations on the skin, called nevi or moles, need careful observation. Some of them can be malignant, so regular checkups with a dermatologist are essential. It's especially important to pay attention to growths that change color or shape. Timely diagnosis and monitoring can help prevent the development of serious conditions, such as melanoma. Consult a specialist at any skin changes to ensure your health and safety.

The app was developed through the collaboration of Russian programmers and scientists. A neural network designed to analyze photographs of moles was trained on a base of five thousand images previously examined by specialists in this field. This helps improve the accuracy of diagnosis and the quality of medical care.

Each image was analyzed by at least two specialists, who determined which moles posed a potential danger and which were normal growths. The neural network achieved a 90% accuracy rate in recognizing oncological diseases, which is comparable to the results of an experienced dermatologist. This confirms the high effectiveness of artificial intelligence in dermatology and its potential for the early diagnosis of skin cancer.

The ProRodinki system consists of several main components. At the center of the system is a neural network hosted on the server, which performs the main data processing. The mobile application serves as a user-friendly interface for users, allowing them to interact with the system. Please note that a stable internet connection is required for the application to function properly.

Each image undergoes a rigorous double check: first, it is analyzed by the neural network, and then by a specialist. In situations where the algorithms cannot provide a clear result, doctors are involved for the final decision. Their expertise not only assists in decision-making but also contributes to the further training of the neural network, improving its efficiency and accuracy in the future.

The application's development lasted three years, during which rigorous testing was conducted. As a result of testing, the system successfully helped doctors diagnose over 1,000 cases of suspicious skin diseases. This application has significantly improved the diagnostic process, providing medical professionals with reliable tools for identifying various skin pathologies.

It is important to understand that the application is not intended for self-treatment. The recommendations provided in "ProRodinki" are probabilistic in nature and are not a substitute for professional medical advice. This app serves as a tool for raising awareness of your health and the need for careful attention to it. Take care of yourself and consult a specialist if you have any doubts.

In the near future, the developers intend to expand the app's functionality by introducing the ability to search for nearby medical facilities and make online appointments with specialists. These innovations will provide users with more convenient access to medical services and save time searching for the necessary specialists. Improved app functionality is aimed at improving the quality of service and simplifying interactions with medical institutions, which will make the process of obtaining medical care more efficient.

Detecting eye diseases using technology

The CRADLE White Eye Detector mobile app is a modern tool for diagnosing eye diseases using photo analysis. Thanks to built-in artificial intelligence, the app effectively identifies dangerous symptoms such as leukocoria—a white glow in the pupil—which can indicate serious illnesses, including retinoblastoma and cataracts. This app significantly simplifies the early diagnosis process, allowing parents and doctors to quickly respond to potential threats to children's vision.

Developed by a team of specialists at Baylor University in the United States, the program aims to diagnose and treat vision problems in children. Unlike adults, who are more likely to recognize visual impairments, children may not notice that they are experiencing difficulties. Failure to promptly detect and treat eye diseases in children can lead to serious consequences, including vision impairment and learning disabilities. It is important to pay attention to children's eye health and conduct regular examinations to detect possible disorders in the early stages.

When using a flash in low-light conditions, a red-eye effect may occur. This phenomenon is caused by light reflection from the blood vessels in the retina and is common and normal. However, if white or yellow eyes appear in photographs, this may signal serious medical issues, such as tumors or other eye diseases. It is important to pay close attention to such signs and promptly consult a specialist for diagnosis and treatment. The app's development began with a personal tragedy. The son of Baylor University professor Brian Shaw was diagnosed with retinoblastoma, which, unfortunately, was made too late, leading to vision loss in one eye. While reviewing old photographs, he noticed leukocoria, which became the impetus for the creation of an app capable of detecting such symptoms early. This app aims to facilitate the early diagnosis of eye diseases in children, which may help prevent similar tragedies in the future.

Professor Shaw challenged his colleagues to develop an app that could automatically analyze photographs of children to detect signs of a white glow in the eyes known as leukocoria. The algorithm, dubbed CRADLE (Computer-Assisted Detector of Leukocoria), was first presented in 2014 and has been continuously improved since then. This app could become an important tool in the early diagnosis of diseases, allowing parents and doctors to promptly respond to potential eye health problems in children.

To train the algorithm, 53,000 images of 40 children were used, including both those with normal vision and those with identified problems. Parents agreed to provide photographs of their children to support the research. Specialists manually annotated the images, noting the presence or absence of leukocoria. This research aims to improve the diagnosis of eye diseases in children and develop more accurate algorithms for the early detection of vision problems.

The app uses ten neural networks to analyze eyes in photographs and transmit data to each of them. Each network is trained in various leukocoria recognition methods, and the final result is formed based on voting between the networks. This approach significantly increases the accuracy of diagnosis, which allowed the app to correctly identify leukocoria in 80% of children with this diagnosis.

To effectively detect symptoms, the algorithm needs to analyze several photographs of the same child. The app scans all images on the smartphone, which significantly increases the chances of early detection of potential problems. If warning signs are detected, the app recommends consulting a doctor for further examination. Using this approach contributes to a more accurate diagnosis and timely intervention, which is important for the child's health.

The app operates offline, allowing users to maintain the privacy of their data, as photos are not sent to the server without their consent. This aspect is especially important for users who care about their privacy. At the same time, the developers emphasize that the application is not intended to replace a consultation with a doctor, and its use does not eliminate the need for professional medical care.

CRADLE White Eye Detector is available for free download on devices with Android and iOS operating systems.