Fujifilm and Kyoto University jointly developed AI-based diagnostic support technology for interstitial pneumonia*1
April 9, 2019
Tokyo, April 9, 2019 — FUJIFILM Corporation (President: Kenji Sukeno) and Department of Respiratory Medicine, Graduate School of Medicine, Kyoto University (Professor Toyohiro Hirai) have succeeded in jointly developing technology that uses artificial intelligence (AI) to automatically categorize and quantify lesions of interstitial pneumonia at an advanced level of precision. Fujifilm strives to introduce a diagnostic imaging support function that can incorporate this technology to its system solutions for clinical use starting from Japan by the end of fiscal year ending March 2021.
The lungs inspire oxygen and expire carbon dioxide through pulmonary alveoli. Interstitial pneumonia is a disease that develops when inflammation or damage to pulmonary alveolar wall causes interstitium tissue to thicken and harden. Interstitial pneumonia can be divided into those due to a clearly-defined cause, such as collagenosis*2 like rheumatoid arthritis, pneumoconiosis, drug-induced pneumonia, and those without specific cause which is called idiopathic interstitial pneumonia (IIP)*3. IIP is a designated intractable disease that cannot be easily treated, and is classified into several types including idiopathic pulmonary fibrosis (IPF)*4. Chest CT scan is one of the most useful examinations for diagnosis of IPF. However, CT images in IPF often show complex and various kinds of abnormal shadows, making it difficult to confirm the diagnosis especially at an early stage of the disease. In some cases, doctors visually observe the changes in individual lung lesions on CT images to confirm diagnosis and choose a treatment approach. In addition, it is known that IPF shows gradual changes in lesions on CT images as the disease progresses, and also sometimes a drastic change called “acute exacerbation” which leads to poor prognosis. It is important to detect signs of acute exacerbation at an early stage.
Furthermore, recently anti-fibrotic drugs for IPF are clinically available, and doctors' needs for quantitative evaluation of the treatment using CT images are growing.
The technology that has been developed this time is AI-based software capable of identifying bronchi, blood vessels, and normal lungs in lung field*5, and seven types of lesions such as reticular opacities, ground-glass opacities and honeycomb lungs*6, and automatically categorizing and measuring them to quantify lesions of interstitial pneumonia. It also divides lung field into 12 zones*7 and shows the volume and ratio of lesions for each of the zones so that clinicians can examine the distribution and progression of lesions in lung field in details.
In the spring of 2018, Fujifilm started joint research on developing this technology in partnership with Professor Toyohiro Hirai at Department of Respiratory Medicine, Graduate School of Medicine, Kyoto University. The AI technology, developed by Fujifilm, to categorize and quantify lesions of interstitial pneumonia, was applied to case data held by Kyoto University to repeatedly evaluate its identification performance and issue feedback for improvement. Variation of image patterns representing each of lesion characteristics was analyzed for further improvement before establishing the novel technology with identification performance of advanced precision.
<Comments by Professor Toyohiro Hirai at Department of Respiratory Medicine, Graduate School of Medicine, Kyoto University>
This novel technology to categorize and quantify various types of abnormal opacities found on chest CT images in the patients with interstitial pneumonia, has the potential of numerous clinical applications as follows:
- Assisting image diagnosis of interstitial pneumonia
- Quantifying changes in opacities on CT images for objective evaluation of the disease progress
- Detailed assessment of the patient's condition by examining lesions in each of the 12 lung zones
- Objective and quantitative evaluation of the effectiveness of treatment
- Applying for novel indicators to evaluate efficacy of the treatment in clinical trial of new drugs
- Applying for clinical researches to investigate the pathophysiology and predict prognosis of interstitial pneumonia
Fujifilm has been working on developing AI technology that can be used for assisting medical diagnostic imaging, facilitating workflow at the medical frontline and delivering maintenance services for medical equipment. AI technology for use in these domains will be marketed under the brand name, “REiLI.” In order to deliver solutions that meet various needs and workflows at the medical frontline, Fujifilm will continue to make in-house technological development, while partnering with Japanese and overseas AI vendors with advanced technologies to achieve fast-paced development of solutions aimed at assisting doctors in diagnostic imaging and streamlining their workflow.
(1) Automatically identifying lungs with interstitial pneumonia, which have structural characteristics different from normal lungs, at a high level of accuracy
(2) Identifying bronchi, blood vessels, normal lungs and seven types of lesions such as reticular opacities, ground-glass opacities and honeycomb lungs, and automatically categorizing and quantifying lungs in CT images to display
(3) Comparing past images (A below) and latest images (B below) of the same patient, and displaying quantified values in graphs for comparison by lesion characteristics or zones
- A and B shows CT image's axial view (transverse plane), sagittal view (longitudinal plane), coronal view (vertical plane into ventral and dorsal) and 3D images, allowing users to observe chronological changes and make comparison.
- C and D in the above image show that the volume of normal lungs has decreased from the past in both right and left lungs as well as in all zones. These graphs can be switched to graphs showing comparison by lesion characteristics and zones.