Significance:
Lung cancer is by far the leading cause of cancer deaths, accounting for 1 in 5 of all cancer deaths worldwide, with more than 1.8 million deaths in 2020. Even with improving treatment options, 66% of lung cancers are diagnosed at a late stage. Lung nodules are small abnormal areas surrounded by normal lung parenchyma. They are discovered on chest computed tomography (CT) images in approximately 1.6 million people annually in the US. Most nodules are not cancer, often the result of old infections or scar tissue, but nodule growth or other concerning features require a biopsy to rule out cancer. Accurate assessment and management of lung nodules are crucial.
Current Approaches:
Bronchoscopy is the preferred diagnostic technique to evaluate lung nodules that are suspicious for lung cancer. Conventional bronchoscopy has a low complication rate but is limited to central lesions and has a low diagnostic yield (15%-31%). CT-guided transthoracic needle biopsy (TTNB) has a high diagnostic yield (67-97%) but is limited to peripheral lesions and has a high complication rate.
Recent advances in (1) bronchoscopy navigation techniques and (2) intra-operative imaging have enabled physicians to safely navigate within the lung and sample lung lesions with increasing accuracy. These techniques guide the bronchoscopist by creating a virtual pathway to a target lesion, but limitations in the accuracy remain.
Bronchoscopy is the preferred diagnostic technique to evaluate lung nodules that are suspicious for lung cancer. Conventional bronchoscopy has a low complication rate but is limited to central lesions and has a low diagnostic yield (15%-31%). CT-guided transthoracic needle biopsy (TTNB) has a high diagnostic yield (67-97%) but is limited to peripheral lesions and has a high complication rate.
Recent advances in (1) bronchoscopy navigation techniques and (2) intra-operative imaging have enabled physicians to safely navigate within the lung and sample lung lesions with increasing accuracy. These techniques guide the bronchoscopist by creating a virtual pathway to a target lesion, but limitations in the accuracy remain.
Problem:
Navigation:
Navigational techniques enable the creation of a three-dimensional (3D) virtual map of the airways from a pre-procedural ‘planning’ CT scan. Planning CTs are typically acquired at full inspiration breath hold to acquire a robust accurate three-dimensional mapping of the bronchial tree. A major limitation to all current guided bronchoscopy systems is the reliance on this static pre-procedural ‘planning’ CT scan. Changes in lung anatomy between the planning CT and time of procedure can lead to a discrepancy between the expected and actual location of the lesion. This is known as “CT-to-body divergence” (CTBD). CTBD increases risk and decreases the diagnostic yield and has been reported in up to 30% of navigation bronchoscopy cases.
Navigation:
Navigational techniques enable the creation of a three-dimensional (3D) virtual map of the airways from a pre-procedural ‘planning’ CT scan. Planning CTs are typically acquired at full inspiration breath hold to acquire a robust accurate three-dimensional mapping of the bronchial tree. A major limitation to all current guided bronchoscopy systems is the reliance on this static pre-procedural ‘planning’ CT scan. Changes in lung anatomy between the planning CT and time of procedure can lead to a discrepancy between the expected and actual location of the lesion. This is known as “CT-to-body divergence” (CTBD). CTBD increases risk and decreases the diagnostic yield and has been reported in up to 30% of navigation bronchoscopy cases.
CTBD has many potential contributing factors and is a pervasive problem across all current guided bronchoscopy platforms. Bronchoscopy procedures are dynamic processes, performed in patients who are either breathing spontaneously or with their breathing controlled under sedation. Peripheral lung nodules can move more than 5cm between full inhalation and full expiration. Discrepancies in patient pose also limit the diagnostic yield; for example, planning CTs are typically taken with the arms above the head to minimize imaging dose yet arms can be required to be by one’s sides to avoid muscle damage during long procedures. Changes in table-top (curved versus flat), bed position and immobilization devices can all cause changes to the thoracic anatomy. Planning CTs may be acquired days or weeks before a procedure, and a growing or shrinking lesion can cause changes to a patient’s breathing. Ideally, planning CTs would be performed on the day of treatment, but this can lead to higher health care costs and increased imaging radiation doses due to repeat scans. Even with same day scans, anatomical changes due to partial or total lung collapse (atelectasis) during the procedure are a common side effect of the procedure and can occur within minutes of general anesthesia induction.
Imaging:
Given the need for accurate imaging and guidance to enhance the diagnostic yield further, intra-procedural cone-beam CT (CBCT) has emerged as a method to provide intra-procedural 3D imaging that can improve all phases of bronchoscopy; navigation, targeting and tissue acquisition. Intra-operative CBCT has similar spatial resolution as diagnostic Multi-Slice CT (MSCT) but suffers from reduced contrast resolution and non-calibrated gray scale values, limiting the use of a standardized ‘lung window’ as is used with MSCT and introducing inaccuracies in image registration results when used for navigation. Additionally, CBCT is inherently more susceptible to several types of image artifacts relative to MSCT, primarily resulting from increased scatter radiation with flat panel detectors, a significant issue in the presence of bronchoscopy and biopsy surgical tools which cause streak artifacts in CBCT images. Due to the long acquisition time (>60s), CBCT is particularly susceptible to image blurring due to motion artifacts from both breathing and cardiac motion. Improved algorithms to model and compensate for motion are required to enhance the diagnostic yield of intra-operative image guided peripheral bronchoscopy.
Given the need for accurate imaging and guidance to enhance the diagnostic yield further, intra-procedural cone-beam CT (CBCT) has emerged as a method to provide intra-procedural 3D imaging that can improve all phases of bronchoscopy; navigation, targeting and tissue acquisition. Intra-operative CBCT has similar spatial resolution as diagnostic Multi-Slice CT (MSCT) but suffers from reduced contrast resolution and non-calibrated gray scale values, limiting the use of a standardized ‘lung window’ as is used with MSCT and introducing inaccuracies in image registration results when used for navigation. Additionally, CBCT is inherently more susceptible to several types of image artifacts relative to MSCT, primarily resulting from increased scatter radiation with flat panel detectors, a significant issue in the presence of bronchoscopy and biopsy surgical tools which cause streak artifacts in CBCT images. Due to the long acquisition time (>60s), CBCT is particularly susceptible to image blurring due to motion artifacts from both breathing and cardiac motion. Improved algorithms to model and compensate for motion are required to enhance the diagnostic yield of intra-operative image guided peripheral bronchoscopy.
Solution:
To eliminate CTBD and improve intra-operative image guidance, the following three components are essential. (1) An accurate model of a patient’s thoracic anatomy that accounts for changes in pose and updates in real time to account for respiratory motion. (2) Fast and accurate image registration that accurately accounts for large scale anatomical motions and atelectasis. (3) Direct visualization of bronchoscopy tools, biopsy probes, as well as target lesions and organs-at-risk with a low dose intraoperative imaging solution that is accurate, high resolution and robust in the presence of metallic artifacts.
We have developed an artificial intelligence (AI)/ computer vision (CV) image registration technique using a novel patient-specific skin mesh model that combines a skeletal-pose machine-learning model with state-of-the-art computer vision techniques to enable marker-less and non-invasive real-time 3D human anatomy tracking. We combine this with a patient-specific biomechanical breathing motion model and an advanced 2D grid Cone-beam CT (CBCT) imaging system allows near-CT quality interoperative real-time 4D imaging, provides accurate HU values, and eliminates CT-to-body divergence issues to improve the diagnostic yield of peripheral bronchoscopy.
