Skip to content

3D Optical Scan

Leong L, Wong M, Piazza M, Garry S, Heymsfield S, Shepherd J. Creating Accurate Representations of DXA Scans from 3D Optical Body Surface Scans for Arbitrary Regional Body Composition Analysis. 2021.

Advancing the assessment of human health through technology

AI PHI is located at the University of Hawai’i Cancer Center in Honolulu. We use advanced technology including AI, machine learning, and deep learning to assess human health and predict the risk of diseases such as cancer.

Our Mission

AI-Predicted DXA

Leong L, Wong M, Piazza M, Garry S, Heymsfield S, Shepherd J. Creating Accurate Representations of DXA Scans from 3D Optical Body Surface Scans for Arbitrary Regional Body Composition Analysis. 2021.

Our Directors

John Shepherd

Co-director
Managing and Imaging Sciences

Peter Sadowski

Co-Director
Machine Learning
Meet Our Team

Actual DXA

Leong L, Wong M, Piazza M, Garry S, Heymsfield S, Shepherd J. Creating Accurate Representations of DXA Scans from 3D Optical Body Surface Scans for Arbitrary Regional Body Composition Analysis. 2021.

Research

Our studies use artificial intelligence to improve and quantify the use of digital imaging in the diagnosis of diseases.

Mammography images separated into lipid, water, and protein.
Fig. 2: 3CB, lipid, water, and protein, data, and regions of feature extraction. (2021). Leong LT, Malkov S, Drukker K, Niell BL, Sadowski P, Wolfgruber T, et al. Dual-energy three-compartment breast imaging for compositional biomarkers to improve detection of malignant lesions.
Loss and accuracy curves per epoch of the test and train set of the deep learning network. Best test loss occurred on epoch 482. At that epoch training loss and accuracy were 0.58 and 67.4%, respectively, and test loss and accuracy were 0.499 and 75.2%, respectively. (2019). Hinton B, Ma L, Mahmoudzadeh AP, Malkov S, Fan B, Greenwood H, et al. Deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case study.
Loss and accuracy curves per epoch of the test and train set of the deep learning network. Best test loss occurred on epoch 482. At that epoch training loss and accuracy were 0.58 and 67.4%, respectively, and test loss and accuracy were 0.499 and 75.2%, respectively. (2019). Hinton B, Ma L, Mahmoudzadeh AP, Malkov S, Fan B, Greenwood H, et al. Deep learning networks find unique mammographic differences in previous negative mammograms between interval and screen-detected cancers: a case-case study.
Figure 2: Receiver operating characteristic curves show performance of the deep learning predictor as a yes-no decision between any group and the rest of the two groups. AUC = area under the receiver operating characteristic curve. (2021). Zhu X, Wolfgruber TK, Leong L, Jensen M, Scott C, Winham S, et al. Deep Learning Predicts Interval and Screening-detected Cancer from Screening Mammograms: A Case-Case-Control Study in 6369 Women.
Figure 2: Receiver operating characteristic curves show performance of the deep learning predictor as a yes-no decision between any group and the rest of the two groups. AUC = area under the receiver operating characteristic curve. (2021). Zhu X, Wolfgruber TK, Leong L, Jensen M, Scott C, Winham S, et al. Deep Learning Predicts Interval and Screening-detected Cancer from Screening Mammograms: A Case-Case-Control Study in 6369 Women.
3CB non-overlapping ROIs: Lesion (blue), Ring 1 (red), Ring 2 (green), Ring 3 (purple). Each ring is 2mm in width. (2020). Leong L, Giger M, Drukker K, Kerlikowske K, Joe B, Greenwood H, et al. Three compartment breast machine learning model for improving computer-aided detection.
3CB non-overlapping ROIs: Lesion (blue), Ring 1 (red), Ring 2 (green), Ring 3 (purple). Each ring is 2mm in width. (2020). Leong L, Giger M, Drukker K, Kerlikowske K, Joe B, Greenwood H, et al. Three compartment breast machine learning model for improving computer-aided detection.
ROC melanoma vs non-melanoma 70% trained, 20% validate, 10% test. (2021). Willingham Jr ML, Spencer SY, Lum CA, Sanchez JMN, Burnett T, Shepherd J, et al. The potential of using artificial intelligence to improve skin cancer diagnoses in Hawai‘i’s multiethnic population.
ROC melanoma vs non-melanoma 70% trained, 20% validate, 10% test. (2021). Willingham Jr ML, Spencer SY, Lum CA, Sanchez JMN, Burnett T, Shepherd J, et al. The potential of using artificial intelligence to improve skin cancer diagnoses in Hawai‘i’s multiethnic population.
Previous
Next
Our Publications

Difference

Leong L, Wong M, Piazza M, Garry S, Heymsfield S, Shepherd J. Creating Accurate Representations of DXA Scans from 3D Optical Body Surface Scans for Arbitrary Regional Body Composition Analysis. 2021.

Latest News

Loading RSS Feed
AI PHI News

Featured Research

The background images used on this page come from our 2021 presentation, Creating Accurate Representations of DXA Scans from 3D Optical Body Surface Scans for Arbitrary Regional Body Composition Analysis (Leong L, et al.). Watch the video below for a demonstration.

play-sharp-fill

The AI learns to accurately reconstruct DXA scans after learning the structure and detail from 20K images

Our work involves analyzing many different types of images, including DXA and digital mammography.

More Research