토라 비전 AI

Chest radiographs are the primary tool for diagnosing lung diseases, but early lesions are subtle and have low contrast, making them easy to miss even by experienced specialists [107]. In environments with insufficient medical personnel, it is difficult to read numerous X-rays individually, leading to diagnostic delays or errors.

Some existing chest X-ray AIs were available, but they were trained on general-purpose data, which limited their ability to detect localized subtle abnormalities.

##

Torah Vision AI utilizes self-collected, high-resolution (16-bit) chest X-ray data and detects 14 types of chest diseases simultaneously using a ResNet50-based deep learning model [108]. It includes all lesions defined in the ChestX-ray14 public dataset, such as Cardiomegaly, pneumonia, and nodules, and analyzes detailed grayscale levels of 65,536 steps to capture subtle changes in shading [106][109].

Furthermore, it supports doctors' decision-making as a precise auxiliary tool by connecting AI reading results with relevant clinical findings and providing recommendations [110]. It is designed to be seamlessly integrated into the reading workflow by linking with the Medidata platform, and it focuses on application in the medical field by improving the quality of learning data with the Biovia solution to enhance data management quality [111].

The key is "X-rays with more grayscale information and AI specialized for it," achieving higher sensitivity and lower false alarm rates compared to existing methods [109].

The main customers are medical institutions (B2B) such as hospital radiology departments and examination centers. It is especially useful in health examination centers and tuberculosis screening programs that require large-scale chest X-ray reading, and it is supplied in partnership with PACS vendors. (Not for general consumers)

##

It is pursuing approval from the Korean Ministry of Food and Drug Safety, and can enter the global hospital market upon approval from the US FDA in the future. It can be integrated into public health programs or remote reading services in countries with a high prevalence of pulmonary tuberculosis, and can be expanded into B2G projects.

However, with leading competitors (such as Lunit) existing, the key is how many references it secures both domestically and internationally.

##

It gained attention by winning the Innovation Award at CES, but it is quieter as it is a solution for experts rather than the general public. In the AI medical industry, it is considered a strength to utilize high-resolution data and multi-disease coverage, and there is a positive evaluation that it is "a new horizon in chest image reading" [108].

However, it is pointed out that it needs to prove its clinical performance advantage compared to competing AIs that have already entered the market. Currently, the technical concept is excellent, but it is in the early stages of commercialization, and more feedback from actual users needs to be accumulated.