Faculty Bibliography

OBJECTIVE/UNASSIGNED:This study investigates speech decoding from neural signals captured by intracranial electrodes. Most prior works can only work with electrodes on a 2D grid (i.e., Electrocorticographic or ECoG array) and data from a single patient. We aim to design a deep-learning model architecture that can accommodate both surface (ECoG) and depth (stereotactic EEG or sEEG) electrodes. The architecture should allow training on data from multiple participants with large variability in electrode placements and the trained model should perform well on participants unseen during training. APPROACH/UNASSIGNED:We propose a novel transformer-based model architecture named SwinTW that can work with arbitrarily positioned electrodes, by leveraging their 3D locations on the cortex rather than their positions on a 2D grid. We train both subject-specific models using data from a single participant as well as multi-patient models exploiting data from multiple participants. MAIN RESULTS/UNASSIGNED:The subject-specific models using only low-density 8x8 ECoG data achieved high decoding Pearson Correlation Coefficient with ground truth spectrogram (PCC=0.817), over N=43 participants, outperforming our prior convolutional ResNet model and the 3D Swin transformer model. Incorporating additional strip, depth, and grid electrodes available in each participant (N=39) led to further improvement (PCC=0.838). For participants with only sEEG electrodes (N=9), subject-specific models still enjoy comparable performance with an average PCC=0.798. The multi-subject models achieved high performance on unseen participants, with an average PCC=0.765 in leave-one-out cross-validation. SIGNIFICANCE/UNASSIGNED:The proposed SwinTW decoder enables future speech neuroprostheses to utilize any electrode placement that is clinically optimal or feasible for a particular participant, including using only depth electrodes, which are more routinely implanted in chronic neurosurgical procedures. Importantly, the generalizability of the multi-patient models suggests the exciting possibility of developing speech neuroprostheses for people with speech disability without relying on their own neural data for training, which is not always feasible.

Effective communication hinges on a mutual understanding of word meaning in different contexts. We recorded brain activity using electrocorticography during spontaneous, face-to-face conversations in five pairs of epilepsy patients. We developed a model-based coupling framework that aligns brain activity in both speaker and listener to a shared embedding space from a large language model (LLM). The context-sensitive LLM embeddings allow us to track the exchange of linguistic information, word by word, from one brain to another in natural conversations. Linguistic content emerges in the speaker's brain before word articulation and rapidly re-emerges in the listener's brain after word articulation. The contextual embeddings better capture word-by-word neural alignment between speaker and listener than syntactic and articulatory models. Our findings indicate that the contextual embeddings learned by LLMs can serve as an explicit numerical model of the shared, context-rich meaning space humans use to communicate their thoughts to one another.

Sound structures such as phonemes and words have highly variable durations. Thus, there is a fundamental difference between integrating across absolute time (e.g., 100 ms) vs. sound structure (e.g., phonemes). Auditory and cognitive models have traditionally cast neural integration in terms of time and structure, respectively, but the extent to which cortical computations reflect time or structure remains unknown. To answer this question, we rescaled the duration of all speech structures using time stretching/compression and measured integration windows in the human auditory cortex using a new experimental/computational method applied to spatiotemporally precise intracranial recordings. We observed significantly longer integration windows for stretched speech, but this lengthening was very small (~5%) relative to the change in structure durations, even in non-primary regions strongly implicated in speech-specific processing. These findings demonstrate that time-yoked computations dominate throughout the human auditory cortex, placing important constraints on neurocomputational models of structure processing.

Dural arteriovenous fistulas (dAVFs) can occur as complications after surgical procedures, especially following the resection of meningiomas near the dural sinus. This case report presents a 74-year-old male who developed a recurrent sigmoid dAVF following meningioma resection. Initially treated with transvenous embolization and middle meningeal artery embolization, the dAVF recurred with worsening clinical symptoms. Conventional treatment options, including sinus sacrifice and transarterial embolization, were unsuitable due to the critical role of the patient's dominant right sigmoid sinus in cerebral venous drainage. Consequently, a reconstructive approach was employed using a pipeline embolization device (PED) construct. The PED successfully occluded the dAVF while preserving the function of the sigmoid sinus. A follow-up angiogram confirmed stable occlusion and normalization of intracranial venous drainage. This case underscores the potential of flow diversion as a viable treatment option for dAVFs, particularly in scenarios where preserving venous sinus function is paramount.

The characterization of genetic alterations in tumor samples has become standard practice for many human cancers to achieve more precise disease classification and guide the selection of targeted therapies. Cerebrospinal fluid (CSF) can serve as a source of tumor DNA in patients with central nervous system (CNS) cancer. We performed comprehensive profiling of CSF circulating tumor DNA (ctDNA) in 711 patients using an FDA-authorized platform (MSK-IMPACT™) in a hospital laboratory. We identified genetic alterations in 489/922 (53.0%) CSF samples with clinically documented CNS tumors. None of 85 CSF samples from patients without CNS tumors had detectable ctDNA. The distribution of clinically actionable somatic alterations was consistent with tumor-type specific alterations across the AACR GENIE cohort. Repeated CSF ctDNA examinations from the same patients identified clonal evolution and emergence of resistance mechanisms. ctDNA detection was associated with shortened overall survival following CSF collection. Next-generation sequencing of CSF, collected through a minimally invasive lumbar puncture in a routine hospital setting, provides clinically actionable cancer genotype information in a large fraction of patients with CNS tumors.

Adult polyglucosan body disease (APBD) is a rare autosomal recessive glycogen storage disorder that leads to slowly progressive multi-organ dysfunction in adulthood. A novel disease-specific patient-reported outcome measure was created and administered to assess symptom burden and health-related quality of life (HR-QOL) in APBD. Thirty-six participants between 30 and 79 years of age (83% ≥60 years, 56% male) completed the anonymous questionnaire independently or with a caregiver proxy (75% self-report). Unemployment predicted an 18.3 (95% CI: 2.8, 33.8; p = 0.028) higher composite disease severity score and a 28.8 (95% CI: 8.2, 49.4; p = 0.010) higher composite HR-QOL score. Use of one or more assistive devices also predicted a 29.3 (95% CI: 8.3, 50.4; p = 0.011) higher composite disease severity score and a 41.8 (95% CI: 10.9, 72.8; p = 0.013) higher composite HR-QOL score. Proxy survey completion predicted a 19.4 (95% CI: 4.1, 34.7; p = 0.020) higher composite disease severity score compared to self-report. Older age at survey completion predicted a 27.4 higher composite HR-QOL score (95% CI: 2.5, 52.4; p = 0.039) for participants in their sixties compared to those between 30 and 59 years old. The development of the Adult Polyglucosan Body Disease questionnaire on Symptom burden and health-related Quality of life (APBD-SQ) marks an important stride forward in capturing the patient experience as a tool for disease monitoring and future research.

Familial dysautonomia (FD) is a rare sensory and autonomic neuropathy that results from a mutation in the ELP1 gene. Virtually all patients report gastrointestinal (GI) dysfunction and we have recently shown that FD patients have a dysbiotic gut microbiome and altered metabolome. These findings were recapitulated in an FD mouse model and moreover, the FD mice had reduced intestinal motility, as did patients. To understand the cellular basis for impaired GI function in FD, the enteric nervous system (ENS; both female and male mice) from FD mouse models was analyzed during embryonic development and adulthood. We show here that not only is Elp1 required for the normal formation of the ENS, but it is also required in adulthood for the regulation of both neuronal and non-neuronal cells and for target innervation in both the mucosa and in intestinal smooth muscle. In particular, CGRP innervation was significantly reduced as was the number of dopaminergic neurons. Examination of an FD patient's gastric biopsy also revealed reduced and disoriented axons in the mucosa. Finally, using an FD mouse model in which Elp1 was deleted exclusively from neurons, we found significant changes to the colon epithelium including reduced E-cadherin expression, perturbed mucus layer organization, and infiltration of bacteria into the mucosa. The fact that deletion of Elp1 exclusively in neurons is sufficient to alter the intestinal epithelium and perturb the intestinal epithelial barrier highlights a critical role for neurons in regulating GI epithelium homeostasis.

Classic change blindness is the phenomenon where seemingly obvious changes that coincide with visual disruptions (such as blinks or brief blanks) go unnoticed by an attentive observer. Some early work into the causes of classic change blindness suggested that any pre-change stimulus representation is overwritten by a representation of the altered post-change stimulus, preventing change detection. However, recent work revealed that, even when observers do maintain memory representations of both the pre- and post-change stimulus states, they can still miss the change, suggesting that change blindness can also arise from a failure to compare the stored representations. Here, we studied slow change blindness, a related phenomenon that occurs even in the absence of visual disruptions when the change occurs sufficiently slowly, to determine whether it could be explained by conclusions from classic change blindness. Across three different slow change blindness experiments we found that observers who consistently failed to notice the change had access to at least two memory representations of the changing display. One representation was precise but short lived: a detailed representation of the more recent stimulus states, but fragile. The other representation lasted longer but was fairly general: stable but too coarse to differentiate the various stages of the change. These findings suggest that, although multiple representations are formed, the failure to compare hypotheses might not explain slow change blindness; even if a comparison were made, the representations would be too sparse (longer term stores) or too fragile (short-lived stores) for such comparison to inform about the change.