Originally published on BioMed Taiwan Startup Hub, written by the editorial team  (Interview by: Shen Rui AI Founder / Distinguished Professor Chen Zhen-Yu, Taipei Medical University | Written by: Jiang Shi-Qi)

Low-dose computed tomography (LDCT) has become the global standard for lung cancer screening. However, LDCT generates hundreds of images, creating a significant interpretation burden for radiologists and clinicians. The DeepLung detection technology, developed by Professor Chen Zhen-Yu’s team, not only enables rapid lung nodule screening but also utilizes the same scan data to assess coronary artery calcification scores, emphysema, and bone density. By efficiently addressing common age-related disease risks in a single screening, this innovation enhances diagnostic accuracy and efficiency for physicians while meeting patients’ demand for precision medicine.

Lung cancer has the highest mortality rate worldwide. In response, the Ministry of Health and Welfare launched an early detection program in 2022, offering free low-dose computed tomography (LDCT) scans to two high-risk groups: individuals with a family history of lung cancer and heavy smokers. This initiative expanded the existing cancer screening program—which previously covered cervical, breast, oral, and colorectal cancers—by adding lung cancer as the fifth screening category.

Compared to full-dose CT scans, low-dose CT imaging provides less information but exposes patients to only one-fifth of the radiation dose, making it suitable for large-scale disease screening. A research team led by Professor Chen Chen-Yu at Taipei Medical University has developed the DeepLung detection technology using LDCT imaging. This technology not only identifies lung nodules but also predicts coronary artery calcification, a feature previously analyzable only with full-dose imaging. With just one click, it generates an automated comprehensive report, significantly reducing the workload of medical personnel and lowering misdiagnosis rates. Moreover, it enhances early diagnosis opportunities for high-risk patients.

The team has utilized LDCT imaging data to develop four diagnostic modules.

DeepLung, led by senior radiologists, was designed specifically for screening purposes, incorporating every detail of actual clinical workflows. Professor Chen Chen-Yu stated that the team’s goal is to enable simultaneous screening of both the lungs and heart, allowing patients undergoing LDCT lung cancer screening to receive additional insights into other critical subclinical conditions, such as coronary artery calcification, emphysema, and osteoporosis.

Using 3D deep learning algorithms, the team has developed multiple models capable of analyzing images at different levels of resolution. To facilitate multi-disease screening, Professor Chen explained that clinical physicians first contribute their expertise in workflow processes and user interface design, guiding medical imaging AI engineers in model training and development. The data is derived from a vast collection of real-world clinical imaging, which, according to Professor Chen, significantly enhances AI model sensitivity and accuracy through effective training.

Currently, DeepLung has established four diagnostic modules: LungRads (lung cancer), CAC (coronary artery calcification score), BMD (bone mineral density), and COPD (emphysema). The lung nodule detection model achieves 86% accuracy at an average of 1 false positive per scan (FPS), increasing to 93% at 2 FPS and 96% at 3 FPS. Meanwhile, the coronary artery calcification model reaches a screening sensitivity of up to 96% for high-risk patients (CAC score >100).

The DeepLung technology, developed by Professor Chen’s team, has successfully built four disease detection modules. The team has also completed an AI platform for physicians to use online, enabling immediate predictions after a patient undergoes an LDCT scan. The platform integrates seamlessly with the National Health Administration’s lung cancer screening program, generating customized reports. After the physician verifies or modifies the data, a report can be finalized and output with a single click in as little as one minute.

DeepLung Undergoing Clinical Trials, Targeting Hospitals, Insurance Industry, and Health Screening Centers

Professor Chen Chen-Yu estimates that approximately 500,000 high-risk individuals in Taiwan qualify for government-funded LDCT screening, requiring regular follow-ups every two years. Additionally, 75% of the sub-healthy population in both domestic and international markets could become potential service users. As a result, DeepLung primarily positions itself within the health screening market, targeting hospitals and health examination centers. For medical institutions offering health screening services, DeepLung significantly enhances the efficiency, quality, and functionality of imaging screenings while expanding screening indicators. This not only increases revenue for health screening providers but also reduces the reporting workload for radiologists.

The DeepLung lung nodule screening product was fully developed in 2023 and is currently undergoing regulatory clinical trials. In the same year, the team established the startup company DeepRad.AI to commercialize their technology. The product is expected to launch in Taiwan by 2024, with plans to apply for regulatory approval from the U.S. FDA and Japan’s PMDA for international expansion.Professor Chen also revealed that the team has completed the prototype development of DeepBrain, an early-stage dementia screening tool, and plans to further develop DeepBreast, a high-dimensional interactive AI platform for breast cancer screening.

Professor Chen Chen-Yu (fifth from the right) and his team were honored with the 18th National Innovation Award in the “Academic-Industrial Innovation – Smart Healthcare & Health Technology” category (Source: BioMed Taiwan).