Devices enabled by artificial intelligence (AI) and machine learning (ML) are being introduced for clinical use at an accelerating pace. In a dynamic clinical environment, these devices may encounter conditions different from those they were developed for. The statistical data mismatch between training/initial testing and production is often referred to as data drift. Detecting and quantifying data drift is significant for ensuring that AI model performs as expected in clinical environments. A drift detector signals when a corrective action is needed if the performance changes. In this study, we investigate how a change in the performance of an AI model due to data drift can be detected and quantified using a cumulative sum (CUSUM) control chart. To study the properties of CUSUM, we first simulate different scenarios that change the performance of an AI model. We simulate a sudden change in the mean of the performance metric at a change-point (change day) in time. The task is to quickly detect the change while providing few false-alarms before the change-point, which may be caused by the statistical variation of the performance metric over time. Subsequently, we simulate data drift by denoising the Emory Breast Imaging Dataset (EMBED) after a pre-defined change-point. We detect the change-point by studying the pre- and post-change specificity of a mammographic CAD algorithm. Our results indicate that with the appropriate choice of parameters, CUSUM is able to quickly detect relatively small drifts with a small number of false-positive alarms.
Deep brain stimulation (DBS) has the potential to improve the quality of life of people with a variety of neurological diseases. A key challenge in DBS is in the placement of a stimulation electrode in the anatomical location that maximizes efficacy and minimizes side effects. Pre-operative localization of the optimal stimulation zone can reduce surgical times and morbidity. Current methods of producing efficacy probability maps follow an anatomical guidance on magnetic resonance imaging (MRI) to identify the areas with the highest efficacy in a population. In this work, we propose to revisit this problem as a classification problem, where each voxel in the MRI is a sample informed by the surrounding anatomy. We use a patch-based convolutional neural network to classify a stimulation coordinate as having a positive reduction in symptoms during surgery. We use a cohort of 187 patients with a total of 2,869 stimulation coordinates, upon which 3D patches were extracted and associated with an efficacy score. We compare our results with a registration-based method of surgical planning. We show an improvement in the classification of intraoperative stimulation coordinates as a positive response in reduction of symptoms with AUC of 0.670 compared to a baseline registration-based approach, which achieves an AUC of 0.627 (p < 0.01). Although additional validation is needed, the proposed classification framework and deep learning method appear well-suited for improving pre-surgical planning and personalize treatment strategies.
Laser interstitial thermal therapy (LITT) is a novel minimally-invasive neurosurgical ablative tool that is par ticularly well-suited for treating patients suffering from drug-resistant mesial temporal lobe epilepsy (mTLE). Although morbidity to patients is lower with LITT compared to the open surgical gold standard, seizure freedom rates appear inferior, likely a result of our lack of knowledge of which mesial temporal subregions are most critical for treating seizures. The wealth of post-LITT imaging and outcomes data provides a means for elucidating these critical zones, but such analyses are hindered by variations in patient anatomy and the distribution of these novel data among multiple academic institutions, each employing different imaging and surgical protocols. Adequate population analyses of LITT outcomes require normalization of imaging and clinical data to a common reference atlas. This paper discusses a method to nonrigidly register preoperative images to an atlas and quantitatively evaluate its performance in our region of interest, the hippocampus. Knowledge of this registration error would allow us to both select an appropriate registration method and define our level of confidence in the correspondence of the postoperative images to the atlas. Once the registration process is validated, we aim to create a statistical map from all the normalized LITT ablation images to analyze and identify factors that correlate with good outcomes.
Responsive neurostimulation (RNS) is a novel surgical intervention for treating medically refractory epilepsy. A neurostimulator implanted under the skull monitors brain activity in one or two seizure foci and provides direct electrical stimulation using implanted electrodes to prevent partial onset seizures. Despite significant successes in reducing seizure frequency over time, outcomes are less than optimal in a number of patients. To maximize treatment efficacy, it is critical to identify the factors that contribute to the variance in outcomes, including accurate knowledge of the final electrode location. However, there is as yet no automated algorithm to localize the RNS electrodes in the brain. Currently, physicians manually demarcate the positions of the leads in postoperative images, a method that is affected by rater bias and is impractical for largescale studies. In this paper, we propose an intensity feature based algorithm that can automatically identify the electrode positions in postoperative CT images. We also validate the performance of our algorithm on a multicenter dataset of 13 implanted patients and test how it compares with expert raters.
The basal ganglia and limbic system, particularly the thalamus, putamen, internal and external globus pallidus, substantia
nigra, and sub-thalamic nucleus, comprise a clinically relevant signal network for Parkinson’s disease. In order to manually
trace these structures, a combination of high-resolution and specialized sequences at 7T are used, but it is not feasible to
scan clinical patients in those scanners. Targeted imaging sequences at 3T such as F-GATIR, and other optimized inversion
recovery sequences, have been presented which enhance contrast in a select group of these structures. In this work, we
show that a series of atlases generated at 7T can be used to accurately segment these structures at 3T using a combination
of standard and optimized imaging sequences, though no one approach provided the best result across all structures. In the
thalamus and putamen, a median Dice coefficient over 0.88 and a mean surface distance less than 1.0mm was achieved
using a combination of T1 and an optimized inversion recovery imaging sequences. In the internal and external globus
pallidus a Dice over 0.75 and a mean surface distance less than 1.2mm was achieved using a combination of T1 and FGATIR
imaging sequences. In the substantia nigra and sub-thalamic nucleus a Dice coefficient of over 0.6 and a mean
surface distance of less than 1.0mm was achieved using the optimized inversion recovery imaging sequence. On average,
using T1 and optimized inversion recovery together produced significantly improved segmentation results than any
individual modality (p<0.05 wilcox sign-rank test).
T1-weighted magnetic resonance imaging (MRI) generates contrasts with primary sensitivity to local T1 properties (with lesser T2 and PD contributions). The observed signal intensity is determined by these local properties and the sequence parameters of the acquisition. In common practice, a range of acceptable parameters is used to ensure “similar” contrast across scanners used for any particular study (e.g., the ADNI standard MPRAGE). However, different studies may use different ranges of parameters and report the derived data as simply “T1-weighted”. Physics and imaging authors pay strong heed to the specifics of the imaging sequences, but image processing authors have historically been more lax. Herein, we consider three T1-weighted sequences acquired the same underlying protocol (MPRAGE) and vendor (Philips), but “normal study-to-study variation” in parameters. We show that the gray matter/white matter/cerebrospinal fluid contrast is subtly but systemically different between these images and yields systemically different measurements of brain volume. The problem derives from the visually apparent boundary shifts, which would also be seen by a human rater. We present and evaluate two solutions to produce consistent segmentation results across imaging protocols. First, we propose to acquire multiple sequences on a subset of the data and use the multi-modal imaging as atlases to segment target images any of the available sequences. Second (if additional imaging is not available), we propose to synthesize atlases of the target imaging sequence and use the synthesized atlases in place of atlas imaging data. Both approaches significantly improve consistency of target labeling.
KEYWORDS: Image segmentation, Thalamus, Image registration, Magnetic resonance imaging, Image resolution, Statistical modeling, Lithium, 3D modeling, Signal to noise ratio, Surgery
Accurate and reliable identification of thalamic nuclei is important for surgical interventions and neuroanatomical studies. This is a challenging task due to their small sizes and low intra-thalamic contrast in standard T1-weighted or T2- weighted images. Previously proposed techniques rely on diffusion imaging or functional imaging. These require additional scanning and suffer from the low resolution and signal-to-noise ratio in these images. In this paper, we aim to directly segment the thalamic nuclei in standard 3T T1-weighted images using shape models. We manually delineate the structures in high-field MR images and build high resolution shape models from a group of subjects. We then investigate if the nuclei locations can be inferred from the whole thalamus. To do this, we hierarchically fit joint models. We start from the entire thalamus and fit a model that captures the relation between the thalamus and large nuclei groups. This allows us to infer the boundaries of these nuclei groups and we repeat the process until all nuclei are segmented. We validate our method in a leave-one-out fashion with seven subjects by comparing the shape-based segmentations on 3T images to the manual contours. Results we have obtained for major nuclei (dice coefficients ranging from 0.57 to 0.88 and mean surface errors from 0.29mm to 0.72mm) suggest the feasibility of using such joint shape models for localization. This may have a direct impact on surgeries such as Deep Brain Stimulation procedures that require the implantation of stimulating electrodes in specific thalamic nuclei.
Deep brain stimulation, which is used to treat various neurological disorders, involves implanting a permanent electrode into precise targets deep in the brain. Accurate pre-operative localization of the targets on pre-operative MRI sequence is challenging as these are typically located in homogenous regions with poor contrast. Population-based statistical atlases can assist with this process. Such atlases are created by acquiring the location of efficacious regions from numerous subjects and projecting them onto a common reference image volume using some normalization method. In previous work, we presented results concluding that non-rigid registration provided the best result for such normalization. However, this process could be biased by the choice of the reference image and/or registration approach. In this paper, we have qualitatively and quantitatively compared the performance of six recognized deformable registration methods at normalizing such data in poor contrasted regions onto three different reference volumes using a unique set of data from 100 patients. We study various metrics designed to measure the centroid, spread, and shape of the normalized data. This study leads to a total of 1800 deformable registrations and results show that statistical atlases constructed using different deformable registration methods share comparable centroids and spreads with marginal differences in their shape. Among the six methods being studied, Diffeomorphic Demons produces the largest spreads and centroids that are the furthest apart from the others in general. Among the three atlases, one atlas consistently outperforms the other two with smaller spreads for each algorithm. However, none of the differences in the spreads were found to be statistically significant, across different algorithms or across different atlases.
In DBS surgery, electrodes are implanted in specific nuclei of the brain to treat several types of movement disorders.
Pre-operative knowledge of the location of the optic tracts may prove useful for pre-operative planning assistance or
intra-operative target refinement. In this article we present a semi-automated method to localize the optic tracts in MR.
As opposed to previous approaches presented to identify these structures, our methods are able to recover the eccentric
shape of the optic tracts. This approach consists of two parts: (1) automatic model construction from manually
segmented exemplars and (2) segmentation of structures in unknown images using these models. The segmentation
problem is solved by finding an optimal path in a graph. The graph is designed with novel structures that permit the
incorporation of prior information from the model into the optimization process and account for several weaknesses of
traditional graph-based approaches. The approach achieved mean and maximum surface errors of 0.35 and 1.9 mm in a
validation study on 10 images. The results from all experiments were considered acceptable.
The use of deep brain stimulation (DBS) for the treatment of neurological movement degenerative disorders requires the
precise placement of the stimulating electrode and the determination of optimal stimulation parameters that maximize
symptom relief (e.g. tremor, rigidity, movement difficulties, etc.) while minimizing undesired physiological side-effects.
This study demonstrates the feasibility of determining the ideal electrode placement and stimulation current amplitude
by performing a patient-specific multivariate optimization using electrophysiological atlases and a bioelectric finite
element model of the brain. Using one clinical case as a preliminary test, the optimization routine is able to find the most
efficacious electrode location while avoiding the high side-effect regions. Future work involves optimization validation
clinically and improvement to the accuracy of the model.
In deep brain stimulation surgeries, stimulating electrodes are placed at specific targets in the deep brain to treat
neurological disorders. Reaching these targets safely requires avoiding critical structures in the brain. Meticulous
planning is required to find a safe path from the cortical surface to the intended target. Choosing a trajectory
automatically is difficult because there is little consensus among neurosurgeons on what is optimal. Our goals are to
design a path planning system that is able to learn the preferences of individual surgeons and, eventually, to standardize
the surgical approach using this learned information. In this work, we take the first step towards these goals, which is to
develop a trajectory planning approach that is able to effectively mimic individual surgeons and is designed such that
parameters, which potentially can be automatically learned, are used to describe an individual surgeon's preferences. To
validate the approach, two neurosurgeons were asked to choose between their manual and a computed trajectory, blinded
to their identity. The results of this experiment showed that the neurosurgeons preferred the computed trajectory over
their own in 10 out of 40 cases. The computed trajectory was judged to be equivalent to the manual one or otherwise
acceptable in 27 of the remaining cases. These results demonstrate the potential clinical utility of computer-assisted path
planning.
A number of groups have reported on the occurrence of intra-operative brain shift during deep brain stimulation (DBS)
surgery. This has a number of implications for the procedure including an increased chance of intra-cranial bleeding and
complications due to the need for more exploratory electrodes to account for the brain shift. It has been reported that the
amount of pneumocephalus or air invasion into the cranial cavity due to the opening of the dura correlates with intraoperative
brain shift. Therefore, pre-operatively predicting the amount of pneumocephalus expected during surgery is of
interest toward accounting for brain shift. In this study, we used 64 DBS patients who received bilateral electrode
implantations and had a post-operative CT scan acquired immediately after surgery (CT-PI). For each patient, the
volumes of the pneumocephalus, left ventricle, right ventricle, third ventricle, white matter, grey matter, and cerebral
spinal fluid were calculated. The pneumocephalus was calculated from the CT-PI utilizing a region growing technique
that was initialized with an atlas-based image registration method. A multi-atlas-based image segmentation method was
used to segment out the ventricles of each patient. The Statistical Parametric Mapping (SPM) software package was
utilized to calculate the volumes of the cerebral spinal fluid (CSF), white matter and grey matter. The volume of
individual structures had a moderate correlation with pneumocephalus. Utilizing a multi-linear regression between the
volume of the pneumocephalus and the statistically relevant individual structures a Pearson's coefficient of r = 0.4123 (p
= 0.0103) was found. This study shows preliminary results that could be used to develop a method to predict the amount
of pneumocephalus ahead of the surgery.
Movement disorders affect over 5,000,000 people in the United States. Contemporary treatment of these diseases
involves high-frequency stimulation through deep brain stimulation (DBS). This form of therapy is offered to
patients who have begun to see failure with standard medical therapy and also to patients for which medical therapy
is poorly effective. A DBS procedure involves the surgical placement, with millimetric accuracy, of an electrode in
the proximity of functional areas referred to as targets. Following the surgical procedure, the implant, which is a
multi-contact electrode is programmed to alleviate symptoms while minimizing side effects. Surgical placement of
the electrode is difficult because targets of interest are poorly visible in current imaging modalities. Consequently,
the process of implantation of a DBS electrode is an iterative procedure. An approximate target position is
determined pre-operatively from the position of adjacent structures that are visible in MR images. With the patient
awake, this position is then adjusted intra-operatively, which is a lengthy process. The post-surgical programming of
the stimulator is an equally challenging and time consuming task, with parameter setting combinations exceeding
4000. This paper reports on the status of the Vanderbilt University DBS Project, which involves the development
and clinical evaluation of a system designed to facilitate the entire process from the time of planning to the time of
programming.
We are developing and evaluating a system that will facilitate the placement of deep brain stimulators (DBS) used to
treat movement disorders including Parkinson's disease and essential tremor. Although our system does not rely on the
common reference system used for functional neurosurgical procedures, which is based on the anterior and posterior
commissure points (AC and PC), automatic and accurate localization of these points is necessary to communicate the
positions of our targets. In this paper, we present an automated method for AC and PC selection that uses non-rigidly
deformable atlases. To evaluate the accuracy of our multi-atlas based method, we compare it against the manual
selection of the AC and PC points by 43 neurosurgeons (38 attendings and 5 residents) and show that its accuracy is submillimetric
compared to the median of their selections. We also analyze the effect of AC-PC localization inaccuracy on
the localization of common DBS targets.
The long term objective of our research is to develop a system that will automate as much as possible DBS implantation procedures. It is estimated that about 180,000 patients/year would benefit from DBS implantation. Yet, only 3000 procedures are performed annually. This is so because the combined expertise required to perform the procedure successfully is only available at a limited number of sites. Our goal is to transform this procedure into a procedure that can be performed by a general neurosurgeon at a community hospital. In this work we report on our current progress toward developing a system for the computer-assisted pre-operative selection of target points and for the intra-operative adjustment of these points. The system consists of a deformable atlas of optimal target points that can be used to select automatically the pre-operative target, of an electrophysiological atlas, and of an intra-operative interface. The atlas is deformed using a rigid then a non-rigid registration algorithm developed at our institution. Results we have obtained show that automatic prediction of target points is an achievable goal. Our results also indicate that electrophysiological information can be used to resolve structures not visible in anatomic images, thus improving both pre-operative and intra-operative guidance. Our intra-operative system has reached the stage of a working prototype that is clinically used at our institution.
Delineation of structures to irradiate (the tumors) as well as structures to be spared (e.g., optic nerve, brainstem, or eyes) is required for advanced radiotherapy techniques. Due to a lack of time and the number of patients to be treated these cannot always be segmented accurately which may lead to suboptimal plans. A possible solution is to develop methods to identify these structures automatically. This study tests the hypothesis that a fully automatic, atlas-based segmentation method can be used to segment most brain structures needed for radiotherapy plans even tough tumors may deform normal anatomy substantially. This is accomplished by registering an atlas with a subject volume using a combination of
rigid and non-rigid registration algorithms. Segmented structures in the atlas volume are then mapped to the corresponding structures in the subject volume using the computed transformations. The method we propose has been tested on two sets of data, i.e., adults and children/young adults. For the first set of data, contours obtained automatically have been compared to contours delineated manually by three physicians. For the other set qualitative results are
presented.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.