Dr. Anup Tuladhar Profile

Dr. Anup Tuladhar

at Univ of Calgary

SPIE Involvement:

Author

Publications (4)

Proceedings Article | 4 April 2022 Presentation + Paper

A comparative analysis of the impact of data distribution on distributed learning with a traveling model for brain age prediction

Raissa Souza, Agampreet Aulakh, Pauline Mouches, Anup Tuladhar, Matthias Wilms, Sönke Langner, Nils Forkert

Proceedings Volume 12037, 1203702 (2022) https://doi.org/10.1117/12.2612728

KEYWORDS: Data modeling, Brain, Neuroimaging, Performance modeling, Machine learning, Data centers, Magnetic resonance imaging, Solid modeling, Medical research, Feature extraction

Read Abstract +

Proceedings Article | 4 April 2022 Presentation + Paper

Lesion-preserving unpaired image-to-image translation between MRI and CT from ischemic stroke patients

Alejandro Gutierrez, Anup Tuladhar, Deepthi Rajashekar, Nils Forkert

Proceedings Volume 12033, 1203317 (2022) https://doi.org/10.1117/12.2613203

KEYWORDS: Magnetic resonance imaging, Computed tomography, Ischemic stroke, Brain, Tissues, Image analysis, Neuroimaging, Data modeling, Process modeling, Artificial intelligence, Realistic image synthesis

Read Abstract +

Proceedings Article | 4 April 2022 Presentation + Paper

Stroke lesion localization in 3D MRI datasets with deep reinforcement learning

Samuel Robertson, Anup Tuladhar, Deepthi Rajashekar, Nils Forkert

Proceedings Volume 12033, 120330M (2022) https://doi.org/10.1117/12.2613039

KEYWORDS: Magnetic resonance imaging, Medical imaging, Ischemic stroke, Image segmentation, Data modeling, Machine vision, Machine learning, Computer vision technology

Read Abstract +

Proceedings Article | 15 February 2021 Presentation + Paper

Understanding privacy risks in typical deep learning models for medical image analysis

Nagesh Subbanna, Anup Tuladhar, Matthias Wilms, Nils D. Forkert

Proceedings Volume 11601, 116010E (2021) https://doi.org/10.1117/12.2582014

KEYWORDS: Data modeling, Visual process modeling, Medical imaging, Image segmentation, Magnetic resonance imaging, Image restoration, Machine vision, Ischemic stroke, Image compression, Databases

Read Abstract +

Deep learning in medical imaging typically requires sensitive and confidential patient data for model training. Recent research in computer vision has shown that it is possible to recover training data from trained models using model inversion techniques. In this paper, we investigate the degree to which encoder-decoder like architectures (U-Nets, etc) commonly used in medical imaging are vulnerable to simple model inversion attacks. Utilising a database consisting of 20 MRI datasets from acute ischemic stroke patients, we trained an autoencoder model for image reconstruction and a U-Net model for lesion segmentation. In the second step, model inversion decoders were developed and trained to reconstruct the original MRIs from the low dimensional representation of the trained autoencoder and the U-Net model. The inversion decoders were trained using 24 independent MRI datasets of acute stroke patients not used for training of the original models. Skull-stripped as well as the full original datasets including the skull and other non-brain tissues were used for model training and evaluation. The results show that the trained inversion decoder can be used to reconstruct training datasets after skull stripping given the latent space of the autoencoder trained for image reconstruction (mean correlation coefficient= 0.49), while it was not possible to fully reconstruct the original image used for training of a segmentation task UNet (mean correlation coefficient=0.18). These results are further supported by the structural similarity index measure (SSIM) scores, which show a mean SSIM score of 0.51± 0.14 for the autoencoder trained for image reconstruction, while the average SSIM score for the U-Net trained for the lesion segmentation task was 0.28±0.12. The same experiments were then conducted on the same images but without skull stripping. In this case, the U-Net trained for segmentation shows significantly worse results, while the autoencoder trained for image reconstruction is not affected. Our results suggest that an autoencoder model trained for image compression can be inverted with high accuracy while this is much harder to achieve for a U-Net trained for lesion segmentation.

View contact details

UPDATE YOUR PROFILE

Is this your profile? Update it now.

Sign into your SPIE.org account

Don’t have a profile and want one?

Create an account on SPIE.org

Keywords/Phrases

Search In:

Publication Years