Using A Semi-Automated Approach to Quantify Unidentified Bright Objects in Neurofibromatosis Type 1 and Linkages to Cognitive and Academic Outcomes

Abstract

Neurofibromatosis type 1 (NF1) is an autosomal dominant neurocutaneous syndrome that affects multiple organ systems resulting in widespread symptoms, including cognitive deficits. In addition to the criteria required for an NF1 diagnosis, approximately 70% of children with NF1 present with Unidentified Bright Objects (UBOs) or Focal Areas of Signal Intensity, which are hyperintense bright spots seen on T2-weighted magnetic resonance images and seen more prominently on FLAIR magnetic resonance images (Sabol et al., 2011). Current findings relating the presence/absence, quantities, sizes, and locations of these bright spots to cognitive abilities are mixed. To contribute to and hopefully disentangle some of these mixed findings, we explored potential relationships between metrics related to UBOs and cognitive abilities in a sample of 28 children and adolescents with NF1 (M=12.52 years; SD=3.18 years; 16 male). We used a novel tool, the Lesion Segmentation Tool (LST), to automatically detect and segment the UBOs. The LST was able to qualitatively and quantitatively reliably detect UBOs in images of children with NF1. Using these automatically detected and segmented lesions, we found that while controlling for age, biological sex, perceptual IQ, study, and scanner, “total UBO volume”, defined as the sum of all the voxels representing all of the UBOs for each participant, helped explain differences in word reading, phonological awareness, and visuospatial skills. These findings contribute to the emerging NF1 literature and help parse the specific deficits that children with NF1 have, to then help improve the efficacy of reading interventions for children with NF1.

1. Introduction

Neurofibromatosis Type 1 (NF1) is a complex autosomal dominant disorder caused by a germline mutation in the NF1 gene [1]. The NF1 gene codes for the protein neurofibromin, which by controlling growth and/or differentiation of neurons, suppresses tumors [2, 3]. Thus, logically, a mutation in the NF1 gene can lead to the formation of neurofibromas, or nerve sheath tumors. Neurofibromas, which form along spinal, peripheral, or cranial nerves, are the defining feature of NF1 [1]. Additionally, NF1 is characterized by café-au-lait spots, axillary and/or inguinal freckling, neurofibromas, Lisch nodules in the irises, optic gliomas, and bone lesions [4]. A relatively common disorder, NF1 affects approximately 1/3,500 individuals [5].

Cognitively, NF1 patients often demonstrate a combination of lower intelligence quotient (IQ) [6], visuospatial impairment [6], language impairment [7], inattention and hyperactivity symptomology as seen in ADHD [8], general executive function deficits [9], and/or social cognitive deficits [10, 11]. Neurobiologically, NF1 patients exhibit brain characteristics including macrocephaly, a higher cumulative volume of all subcortical structures, and/or differences in microstructure integrity and connectivity [11].

Though not part of the official diagnostic criteria for NF1, one of the hallmark characteristics of NF1 is Unidentified Bright Objects (UBOs) [11, 12]. Also known as T2-hyperintensities or Focal Abnormal Signal Intensities, UBOs are abnormally bright areas of the brain and spinal cord seen on T2-weighted magnetic resonance images and seen more prominently on fluid-attenuated inversion recovery (FLAIR) images [3, 13,14]. Indeed, the presence of UBOs in children with NF1 is well documented [e.g. 3, 11, 12, 13, 15–20]. While estimates vary, UBOs can generally be found in approximately 70% of children with NF1 [15]. By comparison, only about 4% of healthy individuals present with UBOs [15]. UBOs are most commonly found in the basal ganglia, brainstem, thalamus, cerebellum, and hippocampus [21]. For most NF1 patients, the number and size of UBOs increase during childhood and then decrease and finally disappear during adolescence [22]. See Figure 1 for an image of a large UBO.

Figure 1:

UBOs. Figure 1 depicts, for one participant, a UBO in a raw scan image (left) and the same UBO colored in red (right).

Previously, UBOs were hypothesized to be hamartomas [23], regions of abnormal myelination [24], or heterotopias [25]. However, DiPaolo and colleagues conducted the first histological examination of UBOs and were finally able to classify them [26]. They found vacuoles between 5–100 micrometers in the myelin sheaths, allowing for the characterization of UBOs as intramyelinic edema. Billiet and colleagues conducted a follow-up study using 3T dMRI; they found evidence of water between myelin layers, supporting the previous findings [27]. Further, neither study found evidence of demyelination and/or inflammation [26, 27]. Therefore, current evidence to date suggests that intramyelinic edema may underlie the presence of UBOs. Because indices of the integrity of white matter (linked to myelin) have been associated with various aspects of cognition [28–30] including reading [31], vocabulary development [32], and executive decision making [33, 34] it follows that the UBOs observed in NF1 might disrupt cognition.

1.1. UBO Detection and Measurement.

UBOs must first be detected and segmented so they can be measured and then potentially related to cognitive processes. A review of the literature indicates that UBOs are most often manually detected and segmented by radiologists, neurologists, or trained researchers [35]. However, this manual detection and segmentation process requires intensive training and is therefore time-consuming and expensive. As a result, tools that can detect, trace, and measure UBOs reliably and automatically are desirable [35]. Indeed, several tools have already been created and are in the processes of being tested and validated. These tools include the FreeSurfer Image Analysis Suite [36], Brain Intensity AbNormality Classification Algorithm (BIANCA) [37], UBO Detector [38] [35], and the Lesion Segmentation Toolbox (LST) [39]. While it is not yet clear which of the several programs available is best for automatic white matter hyperintensity segmentation [35], we selected the LST because it is fully automated, does not require a training dataset, and has been reported as reliable in patients with Multiple Sclerosis (MS) [40]. Of particular importance to the current study, the LST contains several algorithms that may be particularly advantageous for detecting, tracing, and measuring UBOs in NF1: the Lesion Growth Algorithm (LGA) and the longitudinal pipeline. The LGA uses a combination of T1 and FLAIR images to produce a lesion probability map [39]. The longitudinal pipeline compares these lesion probability maps, created using the LGA, across multiple time points to parse true, reliable UBOs from potential artifact. The longitudinal pipeline then produces corrected lesion probability maps for each time point and an overall lesion change plot illustrating those corrections [41]. LST was originally created to segment lesions in patients with MS [42]; it has been used in several papers since [40, 43, 44] and appears to be reliable [40]. Given the similarities in presentation on FLAIR images for MS lesions to NF1 UBOs, it follows that the LST could potentially be used to segment lesions in patients with NF1.

1.2. UBOs and Linkages to Cognition.

Once detected, segmented, and measured, potential relationships between UBOs and cognitive processes can be explored. Some research has found significant relationships between the presence/absence of UBOs and general cognitive abilities in children with NF1. Children with NF1 with UBOs have been found to have lower IQs than children with NF1 without UBOs [17, 21, 45, 46]. Children with NF1 with UBOs also have lower IQs than their typically developing siblings [47]. While the findings regarding UBOs and overall intelligence are likely the most robust, other relationships between UBOs and cognition have also been found. North and colleagues found that children with NF1 with UBOs had lower language scores and impaired visuomotor integration as compared to children with NF1 without UBOs [46]. Supporting these findings, Joy and colleagues found that children with NF1 with UBOs had lower attention scores, decreased visuospatial skills, and decreased executive functioning skills as compared to children with NF1 without UBOs [45]. More recently, Piscitelli and colleagues found that children with NF1 with UBOs had possibly decreased visuospatial skills and decreased language skills, due to their lower IQ, as compared to children with NF1 without UBOs [21].

Emerging research has suggested that the location of the UBOs may also be related to cognition. Presence of UBOs in the thalamus appears to relate to decreased overall intelligence [13, 48–50]. Additionally, Moore and colleagues found that children with NF1 who had UBOs in the thalamus had decreased memory, increased distractibility, and decreased attention as compared to children with NF1 who had UBOs in other brain regions [50]. Goh and colleagues found that children with NF1 who had UBOs in the left globus pallidus had lower attention, and children with NF1 who had UBOs in the right middle cerebellar peduncle had lower sensorimotor skills, all as compared to children with NF1 without UBOs [48]. Chabernaud and colleagues found that children with NF1 who had UBOs in the thalamus and/or striatum had lower visuospatial skills as compared to children with NF1 without UBOs [13].

However, not all research has found significant relationships between presence/absence or location of UBOs and cognitive processes [e.g. 51, 52]. Furthermore, most research conducted on UBOs has used presence/absence of UBOs and only a few have used a more specific metric like one related to UBO volume [47, 48]. Denckla and colleagues used a UBO/brain ratio (total summed volume of UBOs divided by total brain tissue volume) [47] and Goh and colleagues used the sums of the cross-sectional areas of the UBOs, which they labeled as volume [48]. Denckla and colleagues found that this ratio of total UBO volume to total brain tissue volume was not related to the discrepancy in IQs between the unaffected siblings and the siblings with NF1 [47]. Goh and colleagues found that volume of UBOs specifically in the thalamus was directly related to neuropsychological functioning as indexed by Wechsler IQ, Hong Kong List Learning Test, and the Rey-O Complex Figure Test; interestingly, the greater the volume, the better the performance on neuropsychological assessments [48]. Additionally, most research conducted on UBOs has been relative to IQ or broader cognitive processes and not more specific processes like those related to reading or visuospatial processing. Therefore, a different metric of UBO volume and more specific cognitive measures might provide more consistent findings than those that used global cognitive measures such as IQ.

1.3. Present Study.

The present study aims to explore potential relationships between UBOs and the cognitive abilities of children with NF1 while investigating the efficacy of a novel automated lesion segmentation tool. We aimed to answer the following research questions: (1) Can the LST reliably detect and segment UBOs in the scans of children and adolescents with NF1? (2) Does total UBO volume relate to any cognitive skills? And, if UBOs do related to cognitive functions, are these patterns global or related to specific cognitive processes?

2. Methods

2.1. Participants

Analyses were performed on a sample of 28 children and adolescents with NF1 taken from two larger longitudinal intervention studies investigating the efficacies of various reading interventions in children and adolescents with NF1 compared to children and adolescents with reading disabilities and typically developing children and adolescents. Data collection began in 2007 and is ongoing. Research ethics review and approval was obtained through the Johns Hopkins Institutional Review Board and the Vanderbilt Institutional Review Board. Parents of children enrolled in the study provided written consent and children provided verbal assent to participate in the study.

All participants included in these present analyses had a clinical diagnosis of NF1, as confirmed by medical records provided by the participant’s physician prior to study enrollment. Participants were recruited locally and nationally through educational clinics, neuropsychology clinics, neurofibromatosis advocacy and support organizations, and the clinicaltrials.gov website (trial identifier NCT00624234). They were included in the study if they were MRI compatible, had normal or corrected-to-normal vision, had normal hearing, and had no known history of spina bifida, cerebral palsy, traumatic brain injury, and/or brain tumors. Participants were not excluded for ADHD, autism symptomology, or mild to moderate intellectual disability, which are all known to commonly co-occur with NF1. Participants continued medications as usual, including medications for ADHD and depression. One NF participant took medication due to a history of seizures. The average age was 12.52 years, ranging from 8.25 years-17.92 years with a standard deviation of 3.18 years. Sixteen participants were male and 12 participants were female. All participants spoke American English at a level sufficient for school. Of the 28 children in this subsample, 1 was Native American (3.57%), 1 was more than one race (3.57%), 1 preferred to not answer (3.57%), 1 was Asian (3.57%), 3 were Black/African-American (10.71%), and 21 were white (75%).

2.2. Measures

2.2.1. Cognitive Measures.

A battery of cognitive assessments was administered to each participant in the study. While each battery assessed the same constructs underlying reading and cognition, the specific assessments included in each battery differed slightly between the two studies. As per Breen and colleagues’ analyses of assessments assessing the same constructs (e.g. WJ-III and WRMT-III) [53] and manuals’ guidance regarding scores from different editions of the same assessment (e.g. CELF-4 and CELF-5), data were pooled across studies for analyses. Specific scores from some of the subtests can be combined to form composite scores such as perceptual IQ (block design and matrices), verbal IQ (similarities and vocabulary), and full-scale IQ. To account for all of these combinations, all analyses included study as a covariate. Data from these cognitive measures were collected and managed using REDCap electronic data capture tools hosted at Vanderbilt University [54, 55].

• IQ: Study A included the Wechsler Intelligence Scale for Children, Fourth Edition (WISC-IV) [56]. Study B included the Wechsler Intelligence Scale for Children, Fifth Edition (WISC-V) [57] or the Wechsler Abbreviated Scale Intelligence, Second Edition (WASI-II) [58], depending on the participant’s age. The WISC-IV, WISC-V, and WASI-II all assess perceptual (block design and matrices) and verbal (similarities and vocabulary) intelligences.

• Visuospatial Skills: Studies A and B both included the Judgement of Line Orientation (JLO) [59]. The JLO assesses visuospatial skills.

• Working Memory: Study A included the Digit Span subtest from the WISC-IV [56]. Study B included the Digit Span test from the Wechsler Adult Intelligence Scale, Fourth Edition (WAIS-IV) [60] or the Wechsler Intelligence Scale for Children, Fifth Edition (WISC-V) [57], depending on the participant’s age. The Digit Span subtest, a composite of forward and backward digit span assessments, assesses verbal short-term and working memory.

• Language: Study A included the Clinical Evaluation of Language Fundamentals, Fourth Edition (CELF-4) [61]. Study B included the Clinical Evaluation of Language Fundamentals, Fifth Edition (CELF-5) [62]. The CELF-4 and CELF-5 assess language and grammar.

• Phonological Awareness: Study A included the Comprehensive Test Of Phonological Processing, First Edition (CTOPP) [63]. Study B included the Comprehensive Test Of Phonological Processing, Second Edition (CTOPP-2) [64]. The CTOPP and CTOPP-2 assess phonological awareness and processing.

• Word Reading: Study A included the Woodcock-Johnson, Third Edition (WJ-III) [65] and the Test Of Word Reading Efficiency, First Edition (TOWRE) [63]. Study B included the Woodcock Reading Mastery Tests, Third Edition (WRMT-III) [66] and the Test Of Word Reading Efficiency, Second Edition (TOWRE-2) [64], as well as an in-house Word List assessment of single word reading accuracy (Word List A, Word List B). The WRMT-III and WJ-III assess reading and reading-related skills under untimed conditions, whereas the TOWRE and TOWRE-2 measure these skills under timed conditions.

• Passage Reading Fluency: Study A included the Test Of Silent Contextual Reading Fluency, First Edition (TOSCRF) [67]. Study B included the Test Of Silent Contextual Reading Fluency, Second Edition (TOSCRF-2) [68]. The TOSCRF and TOSCRF-2 assess reading fluency.

2.2.2. Imaging.

Magnetic resonance images (MRI) were collected at either Johns Hopkins University’s Kennedy Krieger Institute (KKI) in Baltimore, Maryland through the FM Kirby Research Center or Vanderbilt University’s (VU) Institute of Imaging Science in Nashville, Tennessee through the Human Imaging Core. Although the sample used in present analyses comes from two studies (Study A and Study B), given the activities of the imaging centers, three scanners were used to collect images across the two studies. Images collected at KKI from 2007–2009 were obtained on one scanner, a 3.0 Tesla Philips Achieva MR scanner with an 8-channel headcoil. Images collected at VU from 2009–2013 were obtained on another 3.0 Tesla Philips Achieva MR scanner with an 8-channel headcoil. These two aforementioned scanners were used in Study A. Images collected at VU from 2018-present were obtained on a 3.0 Tesla Philips Ingenia CX MR scanner with a 32-channel headcoil. This aforementioned scanner was used in Study B.

Structural MR images were acquired from each participant using T1-weighted and T2 FLAIR sequences for both studies.

• Study A: For the T1-weighted sequence, we followed the magnetization-prepared gradient recalled echo (MP-RAGE) protocol with 256 × 200 scan acquisition matrix; 200 slices (coronal); 1-mm slice thickness; repetition time (TR) = 7.9ms; echo time (TE) = 3.7s; flip angle = 8°; voxel size = 1-mm³ isotropic; acquisition time = 6m58.6s. For the T2 FLAIR sequence, we used 208 × 115 acquisition matrix; 1.11 × 1.60 × 5.00 mm acquisition voxel; 25 transverse slices; 05mm slice gap; TR 9000ms; TE 120ms; TSE factor 30, 0m54s.

• Study B: For the T1-weighted sequence, we followed the magnetization-prepared gradient recalled echo (MP-RAGE) protocol and also included the following parameters: 256-x-256 scan acquisition matrix; 170 slices in sagittal orientation; 1-mm slice thickness; repetition time (TR) = 8.9ms; echo time (TE) = 4.61ms; flip angle = 8°; voxel size = 1-mm³ isotropic; acquisition time = 4m24.3s. For the T2 FLAIR sequence, we used 352 × 185 acquisition matrix; 0.65 × 0.99 × 4.00 mm acquisition voxel; 28 transverse slices; 1mm slice gap; TR 11000ms; TE 125ms; TSE factor 31; 3m40.0s.

Images were uploaded to and stored in XNAT, operated by the Vanderbilt Center for Computational Imaging at Vanderbilt University [69], and processed in ACCRE, owned and operated by the Advanced Computing Center for Research and Education at Vanderbilt University. To account for potential differences and systematic artifacts in scanners and locations, all analyses include scanner as a covariate.

2.2.2.1. Unidentified Bright Objects.

MRI brain scans were preprocessed using MATLAB (versions 2018a and 2021a) [70, 71]. MR images were skull-stripped using the Statistical Parametric Mapping toolbox (version SPM12, 2020) [72] at a threshold of 0.30 (www.fil.ion.ucl.ac.uk/spm). The T1 and FLAIR images were then run through the Lesion Growth Algorithm (LGA) of the Lesion Segmentation Tool (LST; version 3.0.0, 2019) [39] in order to segment the UBOs.

The images were then viewed in MRIcroGL (version 1.2.20211006, 2021) [73], to locate, identify, and measure the volumes of the UBOs. The quantity of UBOs was found by counting each participant’s number of independent UBOs. The volume of each UBO was calculated by MRIcroGL. The total volume of all UBOs was calculated by summing the individual volumes calculated by MRIcroGL. Similar to the threshold used by Barbier and colleagues, UBOs of volumes less than 2 voxels were deemed likely to be artifact and were consequently excluded from analyses [74]. Twelve outlier UBOs that were either one or two voxels were excluded from analyses. Images were also viewed in MRIcron (version 1.0.2019.0902, 2019) [73] to visualize overlap of images across visits [74].

2.3. Analyses

To analyze these data, we first used the LGA algorithm within the LST to detect and segment UBOs in every individual scan, as LGA is the original base algorithm around which the LST was created. Once detected and segmented, we used MRIcroGL to measure the volume of each UBO. We then used the longitudinal pipeline within the LST to coregister the individual lesion probability maps constructed during the LGA for participants in Study B.

Reliability of the LST was assessed quantitatively and qualitatively. To quantitatively assess reliability, we used R (version 4.1.1) [75] to run paired sample t-tests on the total volumes of UBOs from participants in Study B (n=14). While both Study A and Study B included in the present analyses are longitudinal, their visit structures are different; Study B involves four total scans, with two scans separated by only five days. Given the short interval between these two scans, the UBOs detected at both should be similar in quantity and volume; thus, they were used to comprise a quantifiable reliability substudy. To qualitatively assess reliability, we used MRIcron to overlay coregistered lesion probability maps from participants in Study B (n=14). Study A only collected FLAIR images at the first visit, so we decided to use Study B, which collected FLAIR images at each visit, to assess reliability.

To explore relations between participants’ total UBO volumes and scores from the cognitive assessments, we used R (version 4.1.1) [75] to run linear regressions. Almost all regression models controlled for age, sex, study, scanner, and perceptual IQ; models for perceptual IQ and full-scale IQ controlled for just age, sex, study, and scanner. Prior to entering variables into the regression models, data were scaled. Given the constraints of working with children, especially children with neurodevelopmental disorders such as NF1, some participants are missing data from some assessments for reasons such as administration errors and time constraints. Additionally, some subtests were only administered in one study or the other, further leaving out more data.

3. Results

Table 1 contains information about the UBOs and cognitive measures for the 28 participants.

Table 1:

Descriptive Statistics.

Measure	N	Mean	Standard Deviation	Range
UBOs
Number of UBOs/participant	28	3.68	8.25	0–44
Total Volume of UBOs/participant	28	561.64	1337.30	0–6706
IQ (WISC-IV, WISC-V, WASI-II)
Perceptual IQ	28	85.00	12.16	66–123
Verbal IQ	28	88.36	11.28	59–116
Full Scale IQ	28	83.57	8.72	70–109
Word Reading (WRMT-III, WJ-III, Word List)
Letter-Word Identification	28	79.82	11.68	61–115
Word Attack	28	77.21	14.52	55–120
Basic Reading	28	77.82	13.01	57–119
Word List A	14	16.14	2.68	10–20
Word List B	14	17.36	1.65	13–19
Reading Comprehension (WRMT-III, WJ-III)
Passage Comprehension	28	81.86	11.32	68–107
Reading Fluency (TOWRE, TOWRE-2, TOSCRF, TOSCRF-2)
Sight Word Efficiency	27	84.19	13.14	66–132
Phonemic Decoding Efficiency	27	79.96	12.82	60–115
Total Word Efficiency	27	79.81	12.60	63–125
Passage Reading Fluency	27	75.93	9.25	54–96
Language (CELF-4, CELF-5)
Word Classes	18	6.67	3.18	2–13
Formulating Sentences	28	7.29	2.66	1–11
Recalling Sentences	28	7.89	2.50	5–14
Understanding Spoken Paragraphs	11	7.00	2.65	2–10
Semantic Relationships	15	6.93	2.31	2–10
Core Language	28	83.14	12.21	62–111
Phonological Awareness (CTOPP, CTOPP-2)
Elision	28	6.07	2.11	3–10
Blending Words	28	8.79	2.60	5–15
Phoneme Isolation	14	6.43	2.68	2–12
Rapid Digit Naming	27	6.81	3.22	2–14
Rapid Letter Naming	27	6.56	2.86	1–13
Rapid Naming Composite	27	80.26	17.79	49–122
Phonological Awareness Composite	28	83.04	11.07	62–107
Working Memory (WISC-IV, WISC-V, WAIS-IV)
Digit Span	28	7.11	2.28	3–13
Visuospatial Skills (JLO)
Judgement of Line Orientation	28	9.64	6.56	0–25

Table 1 contains descriptive statistics about the UBOs and selected cognitive measures.

First, we assessed reliability of the LST quantitatively and qualitatively. Quantitatively, the UBOs detected at these two visits are not significantly different in quantity with outlier UBOs included or removed (t = −0.583 and p = 0.574; t = 0 and p = 1 respectively). The UBOs detected at these two visits are also not significantly different in volume with outlier UBOs included or removed (t = 0.044 and p = 0.970; t = 0.054 and p = 0.958 respectively). Table 2 contains more information about the reliability. Qualitatively, the lesion probability maps for each of the four visits overlap. Figure 2 illustrates the overlap of UBOs across four visits for one participant.

Table 2:

Reliability.

Category	t-value	p-value
Outliers Included
Number of UBOs	−0.583	0.574
Total UBO Volume	0.044	0.970
Outliers Removed
Number of UBOs	0	1
Total UBO Volume	0.054	0.958

Table 2 contains information about reliability (n=14). Outliers are UBOs that are less than or equal to two voxels in volume.

Figure 2:

LST Reliability. Figure 2 illustrates, for one participant, the four individual lesion probability maps for each visit (red, green, blue, yellow) and then all four maps overlaid. Note the overlap in lesions in the overlaid image.

Next, we investigated potential relationships between total UBO volume and cognitive skills using linear regression models. Given the number of regressions run, we applied FDR corrections to the p-values of subtests within constructs. Total volume of UBOs was significantly related to letter-word identification as measured by the WRMT-III and WJ-III (B = −0.508; corrected p = 0.041) and elision as measured by the CTOPP and CTOPP-2 (B = −0.486; corrected p = 0.048). When statistically significantly related, children with higher total UBO volumes had lower scores on these assessments. Total UBO volume was also significantly related to scores on the judgement of line orientation task (B = −0.405; uncorrected p = 0.008). Again, when statistically significantly related, children with higher total UBO volumes had lower JLO scores. Other measures did not meet criteria for significance; all measures tested are included in Table 3. Table 3 contains statistical information about the relationships between total UBO volume and cognitive skills.

Table 3:

UBOs and Cognitive Measures.

Measure	N	Parameter Estimate	Standard Error	t-value	p-value	Corrected p-value
IQ (WISC-IV, WISC-V, WASI-II)
Perceptual IQ	28	0.163	0.215	0.757	0.457	0.457
Verbal IQ	28	−0.345	0.210	−1.649	0.113	0.226
Full Scale IQ	28	0.019	0.233	0.082	0.935	--
Word Reading (WRMT-III, WJ-III, Word List)
Letter-Word Identification	28	−0.508	0.180	−2.822	0.010 **	0.041 **
Word Attack	28	−0.400	0.204	−1.964	0.063	0.084
Basic Reading	28	−0.461	0.191	−2.422	0.025 **	--
Word List A	14	−0.856	0.468	−1.829	0.101	0.101
Word List B	14	−1.273	0.488	−2.609	0.028 **	0.057
Reading Comprehension (WRMT-III, WJ-III)
Passage Comprehension	28	−0.213	0.220	−0.965	0.346	NA
Reading Fluency (TOWRE, TOWRE-2, TOSCRF, TOSCRF-2)
Sight Word Efficiency	27	−0.393	0.211	−1.860	0.078	0.233
Phonemic Decoding Efficiency	27	−0.289	0.231	−1.251	0.225	0.338
Total Word Efficiency	27	−0.397	0.216	−1.833	0.082	--
Passage Reading Fluency	27	−0.079	0.175	−0.452	0.656	0.656
Language (CELF-4, CELF-5)
Word Classes	18	−0.311	0.165	−1.882	0.087	0.218
Formulating Sentences	28	−0.371	0.207	−1.794	0.087	0.218
Recalling Sentences	28	0.160	0.200	0.800	0.433	0.541
Understanding Spoken Paragraphs	11	1.831	1.558	1.175	0.293	0.488
Semantic Relationships	15	0.307	0.690	0.446	0.666	0.666
Core Language	28	−0.142	0.200	−0.707	0.487	--
Phonological Awareness (CTOPP, CTOPP-2)
Elision	28	−0.486	0.171	−2.849	0.010 **	0.048 **
Blending Words	28	0.019	0.185	0.101	0.920	0.920
Phoneme Isolation	14	0.319	0.684	0.467	0.652	0.920
Rapid Digit Naming	27	−0.067	0.212	−0.316	0.755	0.920
Rapid Letter Naming	27	−0.143	0.199	−0.719	0.481	0.920
Rapid Naming Composite	27	−0.106	0.208	−0.511	0.615	--
Phonological Awareness Composite	28	−0.217	0.195	−1.111	0.279	--
Working Memory (WISC-IV, WISC-V, WAIS-IV)
Digit Span	28	−0.415	0.217	− 1.912	0.070	NA
Visuospatial Skills (JLO)
Judgement of Line Orientation	28	−0.405	0.138	−2.945	0.008 **	NA

Table 3 contains information about the relationships between total UBO volume and cognitive skills. Bolded and italicized measures and p-values indicated with a ** are significant at traditional alpha=0.05. Corrected p-values are FDR corrections for multiple comparisons, calculated using subtest scores. Composite scores were excluded from the FDR correction calculations for redundancy’s sake.

4. Discussion

The present analyses provided intriguing insight into a new tool used for automatic UBO detection and adds to the literature on our understanding of how UBOs in NF1 may impact a variety of cognitive outcomes. We first demonstrated that The Lesion Segmentation Tool (LST) is a reliable tool for detecting UBOs in children with NF1. Qualitatively, the UBOs detected by LST at each of the four visits indeed overlap. Quantitatively, there were no significant differences in UBO quantity and volume between two scans separated by five days. This lack of significant differences held true when outlier UBOs less than or equal to 2 voxels are included and when they were excluded, thus raising the confidence that the LST was detecting true findings vs artifacts.

Using the LST to detect UBOs and MRIcroGL to measure them, we discovered several definitively significant relationships between total volume of UBOs and cognitive and reading skills. Single word reading ability, as measured by letter-word identification, significantly inversely related to total UBO volume. Skills underlying reading, including word attack and phonological awareness, also related to total UBO volume, albeit some not reaching traditional levels of significance. For example, phonological awareness as measured by the elision task in the CTOPP/CTOPP-2, which measures the ability to remove phonological segments from spoken words to form other words, significantly and inversely related to total UBO volume. This finding thus supports prior findings that children with NF1 exhibit deficits in phonological awareness [7, 76]. However, no other measures of phonological awareness, such as blending words or phoneme isolation, were found to relate to total UBO volume. Therefore, the type of phonological deficits linked to UBOs and NF1 may be more specific in nature. In sum, there is evidence that word-level reading skills negatively correlate with total UBO volume, with all scores in this domain lower with higher volume.

Most of the studies that analyze the relationships between UBOs and cognitive abilities use broad cognitive assessments. Although those aforementioned studies have inconsistent findings as discussed in the introduction, we were able to replicate some of those relationships using similar assessments. It has been well-documented that children with NF1 experience visuospatial deficits [e.g. 6, 77, 78]. Furthermore, children with NF1 with UBOs experience worse visuospatial deficits as compared to their peers with NF1 who do not have any UBOs [21, 45, 46]. We were able to support and extrapolate those findings here, as we found that children with larger UBO volumes have lower JLO scores.

However, we were unable to completely replicate some broader previous findings particularly related to working memory and IQ. Some researchers have found that children with NF1 experience difficulties with working memory, often measured with subtests from the Weschler intelligence tests for adults and children [79]. We were able to support those findings here, as we found that the inverse relationship between total volume of UBOs and working memory, as measured by the digit span task, approaches traditional levels of significance. It has been relatively well documented that children with NF1 have lower IQ scores than their typically developing peers or siblings [6, 80]. Some researchers have previously found significant relationships between UBOs, especially when located in the thalamus, and IQ [17, 21, 45, 46, 48]. Similar to Denckla and colleagues [47] and unlike Goh and colleagues [48], however, we did not find any significant relationships between UBO volume and IQ, perceptual reasoning IQ, or verbal IQ, as measured by the Weschler intelligence tests for adults and children. It may be that IQ is more impacted by total brain volume than specific volume of UBOs, which may offer more specificity in impairing certain aspects of cognition.

Clearly, findings regarding UBOs, particularly their volumes, and cognitive measures are inconsistent. These inconsistent findings could potentially be due to any number of reasons including but not limited to: small sample sizes not having enough power to detect significant relationships; differences in neuropsychological assessments used, differences in participant inclusion criteria; differences in participant samples (particularly their ages, given the disappearance of UBOs in adolescence); differences in UBO inclusion criteria; differences in sequences used to image UBOs (FLAIR, T2, density proton weighted, etc.); and, the strength of the magnet in the scanner used [16, 81]. Other reasons for these inconsistences could include differences in the UBO metrics (cross-sectional area, volume, etc.) used in analyses and variations in locations of the UBOs included in samples. Strict replication of methods would help make these findings more consistent.

While these findings are interesting and comprise a cohesive story, this present study is not without its limitations. As is common with many studies involving NF1, our sample size is relatively small. While we feel confident in the significant relationships found, some may yet to be uncovered due to the limited statistical power of this small sample size. Additionally, all participants were involved in pharmacological and/or intensive reading interventions, as those interventions were the primary motivations for these two studies. As the studies have yet to conclude, the authors are presently unaware of which participants received which interventions, and so there could be effects of interventions left unaccounted for in the present analyses, particularly the reliability analyses which involve visits conducted during the pharmacological intervention. With regard to cognitive assessments, we combined scores from assessments that assessed the same constructs. While manuals indicate that different versions of tests are highly correlated and previous literature indicates assessments of the same constructs are highly correlated [53], it is not ideal because of, for example, different testing items and standardization protocols unique to each assessment and version. With regard to imaging, scans were collected using three different scanners at two different sites; the older scanners are likely of lower quality, producing images of lower resolution and contrast, obscuring smaller and/or fainter UBOs and thus preventing the LST from detecting them. Additionally, while our findings strongly suggest that UBO detection was not an artifact due to the longitudinal nature of our study (multiple scans), we did not validate the LST using traditional methods of manual detection and segmentation.

Further research on UBOs and cognition is needed. Future research should consider using more traditional methods and a larger number of participants to validate the LST. Furthermore, it should also consider investigating relationships between UBOs and more specific cognitive skills such as reading. Future research should consider comparing findings using LST with findings using other automatic UBO detection tools (FreeSurfer, BIANCA, UBO Detector) to determine which tool(s) is/are the most accurate and reliable [35]. It should also consider parsing UBOs by location, to examine if UBOs in specific brain regions like the basal ganglia and/or thalamus are particularly significantly predictive, as previous literature has found. Finally, future research should also consider looking at connectivity between UBOs.

5. Conclusions

Neurofibromatosis Type 1 is a complex disorder that has widespread neurological effects. More research is needed to understand the neurological underpinnings of the cognitive deficits seen in children with NF1. However, using automated tools to examine the relationships between UBO volumes and cognitive abilities, such as word reading, phonological awareness, and visuospatial skills, can help create and refine more targeted pharmacological and/or academic and/or behavioral interventions. More effective interventions are more likely to significantly help improve academic performance and quality of life in children with NF1.

Highlights:

• Children with Neurofibromatosis type 1 often present with Unidentified Bright Objects (UBOs).

• The Lesion Segmentation Tool can be used to segment UBOs.

• In previous literature, findings linking UBOs to cognition are inconsistent.

• UBO volume relates to word reading, phonological awareness, visuospatial, and language skills.

Abbreviations:

NF1

Neurofibromatosis type 1

UBO

Unidentified Bright Object

MRI

magnetic resonance image/imaging

LST

Lesion Segmentation Tool

LGA

Lesion Growth Algorithm

FLAIR

fluid-attenuated inversion recovery

ADHD

attention deficit/hyperactivity disorder

MS

Multiple sclerosis

dMRI

diffusion magnetic resonance imaging

IQ

intelligence quotient

WISC-IV

Wechsler Intelligence Scale for Children, Fourth Edition

WISC-V

Wechsler Intelligence Scale for Children, Fifth Edition

WASI-II

Wechsler Abbreviated Scale Intelligence, Second Edition

JLO

Judgement of Line Orientation

WAIS-IV

Wechsler Adult Intelligence Scale, Fourth Edition

CELF-4

Clinical Evaluation of Language Fundamentals, Fourth Edition

CELF-5

Clinical Evaluation of Language Fundamentals, Fifth Edition

CTOPP

Comprehensive Test Of Phonological Processing, First Edition

CTOPP-2

Comprehensive Test Of Phonological Processing, Second Edition

WJ-III

Woodcock-Johnson, Third Edition

TOWRE

Test Of Word Reading Efficiency, First Edition

WRMT-III

Woodcock Reading Mastery Tests, Third Edition

TOWRE-2

Test Of Word Reading Efficiency, Second Edition

TOSCRF

Test Of Silent Contextual Reading Fluency, First Edition

TOSCRF-2

Test Of Silent Contextual Reading Fluency, Second Edition

MP-RAGE

magnetization-prepared gradient recalled echo

TR

repetition time

TE

echo time

TSE

turbo spin echo