(From left) Kim Tae-kyu, DEEPNOID co-founder and executive director, and Hyun Ji-hoon, head of the DEEPNOID research institute, meet with ChosunBiz on Jan. 22 at the DEEPNOID office in Guro-gu, Seoul, and say, “DEEPNOID is launching the first generative AI medical solution in the domestic industry” and “We aim to expand our global business and leap into the medical sector as the OpenAI of healthcare.” /Courtesy of DEEPNOID

As the competition in artificial intelligence (AI) among major corporations worldwide intensifies, medical AI is also evolving rapidly. AI company DEEPNOID has developed a model that automatically generates reading reports, a primary task for radiologists, by implementing generative AI.

Existing medical AI assists in diagnosis by quickly analyzing patients' chest X-ray images and indicating areas of potential disease and their severity. Going a step further, it writes reading reports in real-time, which physicians used to manually record. DEEPNOID plans to commercialize this technology domestically for the first time this year and aim for entry into the global healthcare market.

Kim Tae-kyu, co-founder and executive director of DEEPNOID, met with ChosunBiz on the 22nd in DEEPNOID's office in Guro, Seoul, and noted, "We will obtain approval for the chest X-ray diagnosis and reading model 'M4CXR' based on generative AI by the end of this year and commercialize it." He added, "We are currently conducting clinical research for domestic approval."

DEEPNOID, established in 2008, is an AI specialist company providing AI solutions for various fields, including healthcare (hospitals), security (airports and ports), and IT manufacturing inspection equipment. It was listed on the KOSDAQ through a technology exception in 2021 due to its recognized technological capabilities.

The company's medical AI products include 'DEEP:NEURO,' which analyzes MRA medical images to diagnose cerebrovascular diseases, 'DEEP:CHEST,' which detects multiple chest diseases in real-time from X-rays, and 'DEEP:LUNG,' which detects lung nodules based on computed tomography (CT). All these products have received approval from the Ministry of Food and Drug Safety and are actively used in local hospitals.

DEEPNOID develops the generative AI diagnosis and interpretation solution M4CXR. M4CXR identifies lesions in patients based on chest X-rays and immediately generates an English interpretation on the right side of the screen. /Courtesy of Heo Ji-yoon

Generative AI refers to technology where AI learns from big data and patterns to produce new results in text, images, video, music, and coding, with OpenAI's Chat-GPT being a representative example.

The generative AI solution M4CXR that DEEPNOID is commercializing focuses on automatically generating medical imaging-based disease diagnoses and reading reports, which are primary tasks of specialists. It was developed by training using numerous chest X-ray image datasets and reading report data with large language model (LMM) technology.

Kim said, "Until now, the analysis of images and the written diagnosis had been performed by radiologists, but AI can now automatically generate these, enhancing the efficiency of medical staff and the productivity of hospitals."

Hyeon Ji-hoon, head of the AI research center at DEEPNOID, explained, "Medical AI is becoming increasingly intelligent. Previously, it primarily analyzed frontal chest images, but now it can analyze not only frontal images but also lateral images, and can compare and analyze based on patients' previous visit and imaging diagnosis records."

Chat-GPT also has the capability of analyzing medical imaging diagnoses, and when compared with M4CXR, the performance of M4CXR was found to be superior. The study analyzed diagnostic performance results of M4CXR and Chat-GPT using a random selection of 926 chest X-ray images from Inha University Hospital between 2010 and 2022.

This study focused on key evaluation indicators such as accuracy, false findings, location errors, count errors, and hallucination. M4CXR showed significantly better performance on all these indicators than Chat-GPT. Hallucination refers to the error that generative AI makes when it cannot find the correct answer, bundling similar content from its vast training knowledge to produce an incorrect answer.

Hyeon added, "M4CXR was developed with a focus on reducing the possibility of hallucination errors," stating that, compared to Chat-GPT, it experiences significantly fewer hallucination errors and has superior accuracy. He noted, "The performance of the AI has improved since the time of the study, so the current diagnostic and reading accuracy scores are even higher than those presented in the paper."

He went on to say, "This study demonstrates that M4CXR, which is specialized for healthcare, is superior to the general-purpose Chat-GPT, showcasing distinct advantages. M4CXR is particularly useful in clinical environments, where radiological expertise is limited or excessively required."

DEEPNOID is expanding its generative AI-based solutions starting from the area of chest diseases, also targeting brain and musculoskeletal disorders, while aiming for overseas expansion. In October last year, it established a branch in Dubai, United Arab Emirates, to serve as a base for entering the Middle East and European markets.

The DEEP:NEURO released by the company in 2023 was shown to analyze MRA images 66 minutes faster than radiology residents and 60 minutes faster than radiology specialists in clinical tests involving over 330 patients. This positions it favorably for securing the golden time that determines the fate of stroke patients, making it highly useful in emergency rooms.

Kim noted, "We are preparing for overseas approval to align with the domestic approval timeline for M4CXR," adding, "Our goal is for DEEPNOID to become the OpenAI of the healthcare sector."

The head of the research center added, "Most reading reports in domestic medical settings are written in English already, making it convenient for global launches."

DEEPNOID's annual sales last year are expected to reach 11 billion won, an increase of about 458% compared to the previous year. The cumulative sales for the first three quarters of last year were approximately 8.2 billion won, with a cumulative operating loss of about 7.3 billion won during the same period. Recently, the company's growth has been led by its AI business for the security and IT industries.

Kim said, "By ensuring stable management through the security and IT AI business, we can continuously invest in medical AI development," and added, "Our financial goal this year is to achieve a turnaround to operating profit," while predicting, "The speed of AI adoption and utilization in the medical field will continue to grow globally, ensuring sustained growth for the medical AI business."