Investigating Dopaminergic Abnormalities in Psychosis with Normative Modelling and Multisite Molecular Neuroimaging DOI Creative Commons
Alessio Giacomel, Daniel Martins, Giovanna Nordio

и другие.

medRxiv (Cold Spring Harbor Laboratory), Год журнала: 2023, Номер unknown

Опубликована: Ноя. 27, 2023

Abstract Molecular neuroimaging techniques, like PET and SPECT, offer invaluable insights into the brain’s in-vivo biology its dysfunction in neuropsychiatric patients. However, transition of molecular diagnostics precision medicine has been limited to a few clinical applications, hindered by issues practical feasibility high costs. In this study, we explore use normative modelling (NM) for identify individual patient deviations from reference cohort subjects. NM potentially addresses challenges such as small sample sizes diverse acquisition protocols that are typical studies. We applied two radiotracers targeting dopaminergic system ([ 11 C]-(+)-PHNO [ 18 F]FDOPA) create model groups controls. The models were subsequently utilized on various independent cohorts patients experiencing psychosis. These characterized differing disease stages, treatment responses, presence or absence matched Our results showed exhibited higher degree extreme (∼3-fold increase) than controls, although pattern was heterogeneous, with minimal overlap topology (max 20%). also confirmed value striatal F]FDOPA signal predict response (striatal AUC ROC: 0.77-0.83). Methodologically, highlighted importance data harmonization before aggregation. conclusion, can be effectively after proper harmonization, enabling mechanisms advancing medicine. method is valuable understanding heterogeneity populations contribute maximising cost efficiency studies aimed at comparing cases

Язык: Английский

Style transfer generative adversarial networks to harmonize multisite MRI to a single reference image to avoid overcorrection DOI Creative Commons
Mengting Liu, Alyssa H. Zhu,

Piyush Maiti

и другие.

Human Brain Mapping, Год журнала: 2023, Номер 44(14), С. 4875 - 4892

Опубликована: Июль 20, 2023

Abstract Recent work within neuroimaging consortia have aimed to identify reproducible, and often subtle, brain signatures of psychiatric or neurological conditions. To allow for high‐powered imaging analyses, it is necessary pool MR images that were acquired with different protocols across multiple scanners. Current retrospective harmonization techniques shown promise in removing site‐related image variation. However, most statistical approaches may over‐correct technical, scanning‐related, variation as they cannot distinguish between confounded image‐acquisition based variability population variability. Such methods require datasets contain subjects patient groups similar clinical demographic information isolate the acquisition‐based overcome this limitation, we consider magnetic resonance (MR) a style transfer problem rather than domain problem. Using fully unsupervised deep‐learning framework on generative adversarial network (GAN), show can be harmonized by inserting encoded from single reference image, without knowing their site/scanner labels priori. We trained our model using data five large‐scale multisite varied demographics. Results demonstrated style‐encoding harmonize images, match intensity profiles, relying traveling subjects. This also avoids need control clinical, diagnostic, information. highlight effectiveness method research comparing extracted cortical subcortical features, brain‐age estimates, case–control effect sizes before after harmonization. showed removed variances, while preserving anatomical meaningful patterns. further diverse training set, successfully collected unseen scanners protocols, suggesting promising tool ongoing collaborative studies. Source code released USC‐IGC/style_transfer_harmonization (github.com).

Язык: Английский

Процитировано

25
Recalibrating single-study effect sizes using hierarchical Bayesian models DOI Creative Commons
Zhipeng Cao, Matthew F. McCabe, Peter Callas

и другие.

Frontiers in Neuroimaging, Год журнала: 2023, Номер 2

Опубликована: Дек. 21, 2023

There are growing concerns about commonly inflated effect sizes in small neuroimaging studies, yet no study has addressed recalibrating size estimates for samples. To tackle this issue, we propose a hierarchical Bayesian model to adjust the magnitude of single-study while incorporating tailored estimation sampling variance.

Язык: Английский

Процитировано

3
Style Transfer Generative Adversarial Networks to Harmonize Multi-Site MRI to a Single Reference Image to Avoid Over-Correction DOI Open Access
Mengting Liu, Alyssa H. Zhu,

Piyush Maiti

и другие.

bioRxiv (Cold Spring Harbor Laboratory), Год журнала: 2022, Номер unknown

Опубликована: Сен. 15, 2022

Abstract Recent work within neuroimaging consortia have aimed to identify reproducible, and often subtle, brain signatures of psychiatric or neurological conditions. To allow for high-powered imaging analyses, it is necessary pool MR images that were acquired with different protocols across multiple scanners. Current retrospective harmonization techniques shown promise in removing cross-site image variation. However, most statistical approaches may over-correct technical, scanning-related, variation as they cannot distinguish between confounded image-acquisition based variability population variability. Such methods require datasets contain subjects patient groups similar clinical demographic information isolate the acquisition-based overcome this limitation, we consider MRI a style transfer problem rather than domain problem. Using fully unsupervised deep-learning framework on generative adversarial network (GAN), show can be harmonized by inserting encoded from single reference image, without knowing their site/scanner labels priori . We trained our model using data five large-scale multi-site varied demographics. Results demonstrated style-encoding harmonize images, match intensity profiles, relying traveling subjects. This also avoids need control clinical, diagnostic, information. highlight effectiveness method research comparing extracted cortical subcortical features, brain-age estimates, case-control effect sizes before after harmonization. showed removed variances, while preserving anatomical meaningful patterns. further diverse training set, successfully collected unseen scanners protocols, suggesting promising novel tool ongoing collaborative studies. Source code released USC-IGC/style_transfer_harmonization (github.com)

Язык: Английский

Процитировано

4
Overview of Cancer Management—The Role of Medical Imaging and Machine Learning Techniques in Early Detection of Cancer: Prospects, Challenges, and Future Directions DOI Open Access
Taofik Ahmed Suleiman,

Adetola Mary Tolulope,

Funmilola Wuraola

и другие.

OALib, Год журнала: 2023, Номер 10(04), С. 1 - 21

Опубликована: Янв. 1, 2023

Globally, the advent of new cases cancer has been steadily increasing, with rising mortality and a significant impact on economy.Most malignancy outcomes are linked to early detection, prompt diagnosis, treatment.The need for detection is crucial management.With these increasing numbers, there adoption emerging technologies such as machine learning help improve outcome management.For reasons, in this paper, we reviewed role medical imaging techniques management cancer.In general, technology used generates enormous data hence, can be analysed using output predict potential tumour cells resulting difference cancer.However, despite advantages, some challenges which also discussed review, well recommendations future directions successful utilization management.

Язык: Английский

Процитировано

2
Investigating Dopaminergic Abnormalities in Psychosis with Normative Modelling and Multisite Molecular Neuroimaging DOI Creative Commons
Alessio Giacomel, Daniel Martins, Giovanna Nordio

и другие.

medRxiv (Cold Spring Harbor Laboratory), Год журнала: 2023, Номер unknown

Опубликована: Ноя. 27, 2023

Abstract Molecular neuroimaging techniques, like PET and SPECT, offer invaluable insights into the brain’s in-vivo biology its dysfunction in neuropsychiatric patients. However, transition of molecular diagnostics precision medicine has been limited to a few clinical applications, hindered by issues practical feasibility high costs. In this study, we explore use normative modelling (NM) for identify individual patient deviations from reference cohort subjects. NM potentially addresses challenges such as small sample sizes diverse acquisition protocols that are typical studies. We applied two radiotracers targeting dopaminergic system ([ 11 C]-(+)-PHNO [ 18 F]FDOPA) create model groups controls. The models were subsequently utilized on various independent cohorts patients experiencing psychosis. These characterized differing disease stages, treatment responses, presence or absence matched Our results showed exhibited higher degree extreme (∼3-fold increase) than controls, although pattern was heterogeneous, with minimal overlap topology (max 20%). also confirmed value striatal F]FDOPA signal predict response (striatal AUC ROC: 0.77-0.83). Methodologically, highlighted importance data harmonization before aggregation. conclusion, can be effectively after proper harmonization, enabling mechanisms advancing medicine. method is valuable understanding heterogeneity populations contribute maximising cost efficiency studies aimed at comparing cases

Язык: Английский

Процитировано

0