Umeå University's logo

Objective: Idiopathic normal pressure hydrocephalus (INPH) is a treatable neurological condition, yet predicting which patients will benefit from a cerebrospinal fluid shunt remains challenging. Structural brain MRI is a core part of the diagnostic workup, but traditional radiological measures show limited predictive accuracy. This study aimed to assess whether deep learning and radiomics-based machine learning approaches can provide clinically useful predictions of shunt outcome based on preoperative MRI.

Methods: We investigated 149 shunted INPH patients with available preoperative T1-weighted, T2-weighted, and FLAIR images. Patients were classified as responders (n = 113) or non-responders (n = 36) based on postoperative gait speed improvement. For INPH, this is a large sample with typical outcome distribution. Three artificial intelligence approaches were tested: a late-fusion ensemble of multiple 3D convolutional neural networks; a multimodal intermediate fusion model; and radiomics-based machine learning models trained on features extracted from whole-brain masks. Models were assessed using 10-fold cross-validation. The best performing model on the validation set was selected from each approach. Performance metrics included the area under the receiving operating characteristic curve (AUROC), sensitivity, and specificity.

Results: Performance was considered poor in all models, and none reached an area under the receiving operating characteristic curve above 70%. Of the three methodologies, the best performance was achieved with a radiomics-based model (Linear Discriminant Analysis classifier on T1-weighted images) which achieved an AUROC of 63.7%. In a reduced subset of clearly separated responders and non-responders (n = 72), the best model (late fusion ensemble of 5 convolutional neural networks) reached an AUROC of 69.2%.

Conclusions: Despite the use of advanced artificial intelligence techniques, structural MRI alone were insufficient for reliably predicting gait outcome after surgery in idiopathic normal pressure hydrocephalus. To capture the complexity of the condition and enable clinically meaningful predictions, our findings indicate the need for research investigating multimodal input and using large multi-center datasets.

Idiopathic normal pressure hydrocephalus (INPH) is a neurological disorder characterized by impaired gait and balance, cognitive decline and incontinence, in combination with enlarged lateral ventricles. Although symptoms can often be alleviated through surgical insertion of a cerebrospinal fluid (CSF) shunt, a substantial proportion of patients do not improve after surgery. There is therefore a need for new analytical methods that can extract more informative features from MRI to improve diagnostic and prognostic accuracy.

This thesis consists of the work from four papers with the overall aim to develop and assess artificial intelligence (AI)-based and fully automated MRI-based methods, to improve objective assessment and shunt decision support in INPH.

Several well-known convolutional neural networks (CNNs) were applied to 3D brain magnetic resonance imaging (MRI) data to distinguish between participants with an INPH-typical gait pattern and controls. An ensemble model search was developed to find the optimal ensemble for the task at hand, by optimizing combinations of diverse models. A fusion search strategy was also developed, to determine the optimal fusion points for information fusion between different MRI sequences. Shunt outcome prediction was evaluated with both deep learning approaches, using the two search algorithms, as well as with radiomics-based machine learning models. Finally, a fully automated pipeline was developed for assessment of disproportionally enlarged subarachnoid space hydrocephalus (DESH), utilising image segmentation and image analysis techniques to determine a quantitative DESH metric (qDESH). The work was conducted on brain MRI from one population-based cohort (Paper I), two open access datasets (Paper II), a clinical cohort of shunted INPH patients (Paper III) and a retrospective cohort of INPH patients and controls (Paper IV).

All CNNs distinguished between gait-impaired and controls, in terms of a chi-square test of independence. The optimized ensemble model achieved the highest classification performance, exceeding that of the individual networks and conventional radiological measures. The results support the presence of detectable structural differences in brain MRI between the groups. The sequential search of multimodal fusion points improved classification performance compared with unimodal and conventional fusion strategies, while reducing computational cost. However, when applying these methodologies to predict shunt outcome, no model achieved clinically sufficient performance. These findings indicate that structural MRI alone is not yet reliable for shunt prediction in INPH. The fully automated qDESH pipeline demonstrated high agreement with the established semi-automatic qDESH method, although the agreement was lower than between two raters of the semi-automatic method. The automated measure of qDESH aligned well with visual assessment of DESH.

In conclusion, this thesis advances methodologies for AI-based and automated brain MRI analysis, particularly for INPH. Introducing and evaluating an optimized ensemble strategy, a systematic multimodal fusion approach, and a fully automated quantitative imaging pipeline, the work demonstrates both the potential and the current limitations of advanced and automated MRI analysis in INPH. The fully automated qDESH pipeline showed good agreement with both the semi-automatic method and visual DESH ratings, although further refinement is required before it can be applied in clinical practice. While CNNs can capture differences in brain MRI beyond conventional linear measures, they cannot yet predict shunt response in a clinically useful way. Structural MRI data alone might be insufficient, and additional non-imaging data might be required. The findings highlight the importance of diversity across models and imaging sequences to improve data-driven image assessment. The need for large clinical datasets is a limiting factor, making collaboration among multiple centres necessary to enable further methodological developments. The methodological approaches and insights presented here may also be transferable to other neurological disorders in which MRI plays a central diagnostic role, thereby contributing more broadly to the neuroimaging field.

According to glymphatic system theory, cerebrospinal fluid (CSF) perfuses the brain’s interstitial space to support waste clearance, but the magnitude of this flow and the outflow pathway of interstitial fluid (ISF) in humans remain uncertain. To achieve flow quantification, we applied a compartment-model approach applied in conjunction with serial quantitative MRI data acquired after intrathecal gadolinium administration. Using the method, we estimated CSF-to-ISF inflow to 45 ± 20 mL/h, in patients with suspected idiopathic normal pressure hydrocephalus. Tissue-specific contributions were 34 ± 14 mL/h in cortical gray matter, 11±6 mL/h in white matter, and 0.4 ± 0.3 mL/h in subcortical gray matter, suggesting that CSF perfusion occurs primarily in superficial regions near the subarachnoid space. A lack of correlation between inflow and total craniospinal system outflow (r = 0.03, P = 0.91) suggested that ISF recirculates back into CSF rather than exiting the craniospinal system via a separate route. Independent experiments in healthy older individuals using intravenous gadolinium administration supported ISF-to-CSF recirculation, where contrast material that presumably crossed the blood–brain barrier subsequently appeared in the subarachnoid space, allowing ISF-to-CSF flow quantification. These findings provide a quantitative framework for studying brain clearance in humans and support subarachnoid space recirculation as an important efflux route.

Higher-Level Gait Disorder (HLGD) is a type of gait disorder estimated to affect up to 6% of the older population. By definition, its symptoms originate from the higher-level nervous system, yet its association with brain morphology remains unclear. This study hypothesizes that there are patterns in brain morphology linked to HLGD. For the first time in the literature, this work investigates whether deep learning, in the form of convolutional neural networks, can capture patterns in magnetic resonance images to identify individuals affected by HLGD. To handle this new classification task, we propose setting up an ensemble of models. This leverages the benefits of combining classifiers instead of determining which network is the most suitable, developing a new architecture, or customizing an existing one. We introduce a computationally cost-effective search algorithm to find the optimal ensemble by leveraging a cost function of both traditional performance scores and the diversity among the models. Using a unique dataset from a large population-based cohort (VESPR), the ensemble identified by our algorithm demonstrated superior performance compared to single networks, other ensemble fusion techniques, and the best linear radiological measure. This emphasizes the importance of implementing diversity into the cost function. Furthermore, the results indicate significant morphological differences in brain structure between HLGD-affected individuals and controls, motivating research about which areas the networks base their classifications on, to get a better understanding of the pathophysiology of HLGD.

Multimodal deep learning harnesses diverse imaging modalities, such as MRI sequences, to enhance diagnostic accuracy in medical imaging. A key challenge is determining the optimal timing for integrating these modalities - specifically, identifying the network layers where fusion modules should be inserted. Current approaches often rely on manual tuning or exhaustive search, which are computationally expensive without any guarantee of converging to optimal results. We propose a sequential forward search algorithm that incrementally activates and evaluates candidate fusion modules at different layers of a multimodal network. At each step, the algorithm retrains from previously learned weights and compares validation loss to identify the best-performing configuration. This process systematically reduces the search space, enabling efficient identification of the optimal fusion timing without exhaustively testing all possible module placements. The approach is validated on two multimodal MRI datasets, each addressing different classification tasks. Our algorithm consistently identified configurations that outperformed unimodal baselines, late fusion, and a brute-force ensemble of all potential fusion placements. These architectures demonstrated superior accuracy, F-score, and specificity while maintaining competitive or improved AUC values. Furthermore, the sequential nature of the search significantly reduced computational overhead, making the optimization process more practical. By systematically determining the optimal timing to fuse imaging modalities, our method advances multimodal deep learning for medical imaging. It provides an efficient and robust framework for fusion optimization, paving the way for improved clinical decision-making and more adaptable, scalable architectures in medical AI applications.

Over the last decade, it has become evident that cerebrospinal fluid (CSF) plays a pivotal role in brain solute clearance through perivascular pathways and interactions between the brain and meningeal lymphatic vessels. Whereas most of this fundamental knowledge was gained from rodent models, human brain clearance imaging has provided important insights into the human system and highlighted the existence of important interspecies differences. Current gold standard techniques for human brain clearance imaging involve the injection of gadolinium-based contrast agents and monitoring their distribution and clearance over a period from a few hours up to 2 days. With both intrathecal and intravenous injections being used, which each have their own specific routes of distribution and thus clearance of contrast agent, a clear understanding of the kinetics associated with both approaches, and especially the differences between them, is needed to properly interpret the results. Because it is known that intrathecally injected contrast agent reaches the blood, albeit in small concentrations, and that similarly some of the intravenously injected agent can be detected in CSF, both pathways are connected and will, in theory, reach the same compartments. However, because of clear differences in relative enhancement patterns, both injection approaches will result in varying sensitivities for assessment of different subparts of the brain clearance system. In this opinion review article, the "EU Joint Programme – Neurodegenerative Disease Research (JPND)" consortium on human brain clearance imaging provides an overview of contrast agent pharmacokinetics in vivo following intrathecal and intravenous injections and what typical concentrations and concentration–time curves should be expected. This can be the basis for optimizing and interpreting contrast-enhanced MRI for brain clearance imaging. Furthermore, this can shed light on how molecules may exchange between blood, brain, and CSF.