Gihan P. Ruwanpathirana
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Gihan P. Ruwanpathirana has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Catherine E. Davey
The University of Melbourne
1 shared publication - Christopher Cleon Rowe
Department of Molecular Imaging & Therapy, Austin Health, Heidelberg, Australia
1 shared publication - Colin L. Masters
Florey Institute of Neuroscience and Mental Health
1 shared publication - Leigh A. Johnston
Graeme Clark Institute The University of Melbourne Parkville Victoria Australia
1 shared publication - Petikirige Sadeep Madhushan Thilakarathna
Department of Infrastructure Engineering, The University of Melbourne, Parkville, VIC 3052, Australia
1 shared publication - Robert Williams
The University of Melbourne, Melbourne, VIC Australia
1 shared publication
Latest Publications
- Open Access
- 4 April 2025
- Published by Elsevier in Case Studies in Construction Materials
- Vol. 22
Interpretation and understanding of asphalt crack detection deep learning models using integrated gradient (I.G.) maps
- 4 April 2025
- Published by Elsevier in Case Studies in Construction Materials
- Vol. 22
Open AccessInterpretation and understanding of asphalt crack detection deep learning models using integrated gradient (I.G.) maps
- Open Access
- 16 June 2023
- Published by Wiley in Alzheimer's & Dementia
- Vol. 19 (S1)
Exploring the association between tau‐PET and Aβ‐amyloid Centiloid value using deep learning
Background: The study of how extracellular Aβ‐amyloid (Aβ) drives intracellular, heterogeneous tau accumulation in Alzheimer’s disease has been approached using linear models under an implicit assumption of regionally independent tau accumulation (Lockhart et al, 2017; Iaccarino et al, 2017). Multivariate, non‐linear models which enable relaxation of this assumption can be used; we employed a convolutional neural network (CNN) to identify critical tau topographic changes associated with Aβ‐burden in a cohort with a range of Aβ values, and compared results with linear analyses.Method: 134 subjects with [NAV4694 (Aβ) and [MK6240 (tau) scans were considered. A CNN was trained to map normalised tau (input) to Aβ Centiloid values (CL). CNN models were interpreted using saliency maps, denoting the importance of corresponding voxels to the output. The CNN approach was compared with linear analysis by testing the following voxel‐wise models: estimating tau SUVR from Aβ CL, and saliency from Aβ CL. All models were tested with minimum cluster extent of 200 after controlling for age and sex. Occlusion analysis was used to assess the importance of CNN‐identified salient clusters, which quantified the change in output after removing salient clusters from the input tau images.Result: The Aβ CL values were well predicted by tau topography using the CNN (training R2 = 0.86, validation R2 = 0.75, testing R2 = 0.72). Linear analysis identified widespread correlation between Aβ CL and tau‐SUVR in the brain, while CNN analysis demonstrated focal clusters in frontal lobes, middle cingulate, precuneus, postcentral gyrus and bilateral medial temporal lobes. At low Aβ levels, information from the frontal lobe, middle cingulate and precuneus regions was more predictive of Aβ CL while the information from medial temporal lobes was more predictive of high Aβ levels.Conclusion: The nonlinear data‐driven CNN approach, relaxing the assumption of independent regional tau accumulation, has revealed new associations between tau topography and Aβ burden.
- 16 June 2023
- Published by Wiley in Alzheimer's & Dementia
- Vol. 19 (S1)
Open AccessExploring the association between tau‐PET and Aβ‐amyloid Centiloid value using deep learning
Background: The study of how extracellular Aβ‐amyloid (Aβ) drives intracellular, heterogeneous tau accumulation in Alzheimer’s disease has been approached using linear models under an implicit assumption of regionally independent tau accumulation (Lockhart et al, 2017; Iaccarino et al, 2017). Multivariate, non‐linear models which enable relaxation of this assumption can be used; we employed a convolutional neural network (CNN) to identify critical tau topographic changes associated with Aβ‐burden in a cohort with a range of Aβ values, and compared results with linear analyses.Method: 134 subjects with [NAV4694 (Aβ) and [MK6240 (tau) scans were considered. A CNN was trained to map normalised tau (input) to Aβ Centiloid values (CL). CNN models were interpreted using saliency maps, denoting the importance of corresponding voxels to the output. The CNN approach was compared with linear analysis by testing the following voxel‐wise models: estimating tau SUVR from Aβ CL, and saliency from Aβ CL. All models were tested with minimum cluster extent of 200 after controlling for age and sex. Occlusion analysis was used to assess the importance of CNN‐identified salient clusters, which quantified the change in output after removing salient clusters from the input tau images.Result: The Aβ CL values were well predicted by tau topography using the CNN (training R2 = 0.86, validation R2 = 0.75, testing R2 = 0.72). Linear analysis identified widespread correlation between Aβ CL and tau‐SUVR in the brain, while CNN analysis demonstrated focal clusters in frontal lobes, middle cingulate, precuneus, postcentral gyrus and bilateral medial temporal lobes. At low Aβ levels, information from the frontal lobe, middle cingulate and precuneus regions was more predictive of Aβ CL while the information from medial temporal lobes was more predictive of high Aβ levels.Conclusion: The nonlinear data‐driven CNN approach, relaxing the assumption of independent regional tau accumulation, has revealed new associations between tau topography and Aβ burden.