CHANNELS

Aerospace

Automotive

Automated & Autonomous Vehicles

Battery & Electrification Technology

Defense

Drone TV

Electronics

Manufacturing & Prototyping

Materials

Medical

Motion Control

RF & Microwave Technology

Robotics & Automation

Sensors & IoT

Test & Measurement

Aerospace
Automotive
Automated & Autonomous Vehicles
Battery & Electrification Technology
Defense
Drone TV
Electronics
Manufacturing & Prototyping
Materials
Medical
Motion Control
RF & Microwave Technology
Robotics & Automation
Sensors & IoT
Test & Measurement
Aerospace & Defense
Search Results

How TBP Helps with Skin Lesion Correspondence Localization

Longitudinal tracking of skin lesions is beneficial to the early detection of melanoma. However, it has not been well-investigated in the context of full-body imaging. Watch this video to see a novel framework combining geometric and texture information to localize skin lesion correspondence from a source scan to a target scan in total body photography (TBP).

“Imagine that we want to look for a house in New York City. If we’re given the locations of some landmarks, like the Empire State Building, and how far the house is from those landmarks, we know the approximate region the house is located in,” explains Wei-Lun Huang  , a doctoral candidate in the Department of Computer Science and a member of the Laboratory for Computational Sensing and Robotics’ Biomechanical and Image-Guided Surgical Systems Lab.

Topics:
Medical

More From SAE Media Group

    Transcript

    00:00:00 hi I'm wedong Huang I'm going to present our work in Makai 2023 skin lesion correspondence localization in total body photography it's a joint work between me DaVita Jun Khan Amir Khan's back Michael katzstan and Mary almond tbp is beneficial to the early detection of melanoma the most delicate skin cancer it helps with longitudinal tracking of

    00:00:28 skin lesion which is fighting lesion correspondence and detecting the change of pigmented lesions researchers have built TPP systems and developed algorithms to find Legion respondents however these methods are only applicable to well-controlled environments in particular these methods assume similar body pose and Camera parameters across scans

    00:00:53 the concept of using template mesh for fighting lesion correspondence has been proposed however it's challenging to deform a template mesh to fit varying body shapes additionally geodesically closed locations may be distant from each other in the texture map in a work from Zhao and others they do not use texture in the matching step in response we propose a novel framework

    00:01:20 to violation correspondence based on three key observations first body landmarks endured as a distance can be used to establish an initial course correspondence in 3D measures second texture information can be used to further refine the location of the correspondence lastly similar to how a human annotator will label lesion correspondence once Covenant lesion

    00:01:44 correspondences are found they can be used as new landmarks to help with finding the correspondence for other lesions of Interest the input to our method are source and Target mesh their corresponding landmarks and lesion of interest in the source we would like to find their correspondence in the target first we define geometric feature

    00:02:07 descriptors for each vertex based on the geodesic distance from the vertex to the limericks we then create an initial dense correspondence based on the similarity of depth geometric descriptors measured by normalized cross-correlation while geometric feature descriptors of corresponding vertices on the source and Target mesh are identical when the

    00:02:33 landmarks are in perfect correspondence and when the source and Target mesh differed by an isometry neither of these assumptions holds in real-world data therefore for the initial correspondence we associate a local region with the vertices that are geodesically close to the correspondence and with the furnaces that have high similarity in their geometric descriptors

    00:03:00 then we refine the correspondence based on textured descriptors within the local region specifically we select the echo descriptor defined on the texture signal with three radii the correspondence is derived by the way the sum of the similarity of texture descriptors in three radii while each Source lesion of Interest has a corresponding position on the target

    00:03:24 mesh not all correspondences are localized with high confidence we Define the confidence by texture similarity agreement between geometric and textural correspondences and unique existence of a similar lesion within the region in each iteration the high confidence correspondence will be added as new landmarks we either relatively relax the criteria

    00:03:50 and look for lesion correspondence finally for the remaining low confidence lesion of Interest we find the correspondence by the way the sum of the similarity between their geometric descriptors and texture descriptors we evaluate our method in a public data set skin 3D and a private image reach TPP data set our success rate at a 10 millimeter

    00:04:15 criteria is comparable to the only reported method on the public data set we also demonstrate the effectiveness of the iterative algorithm the iterative method consistently performs better in success rate under the 20 millimeter criteria in conclusion we propose another method to effectively find lesion correspondence across TPP scans through

    00:04:40 3D textured mesh by leveraging geometric and texture information the method is robust to changes in body pose and Camera viewing directions we believe the proposed method will serve as a valuable step for the longitudinal tracking of skin lesions