Earticle

The purpose of this paper is to argue that the theories of language production and comprehension adopt the predictive process of the brain. Predictive processes can be found in many domains of language: discourse, syntactic, semantic, and phonological domains. In the discourse domain, the predictive process is found in the conversations that take turns in about 200 milliseconds. Syntactically anomalous structures, e.g., garden-path sentences, elicit a large positive wave (P600), indicating the unexpectedness of the structure. Semantic prediction is exhibited in an evoked response (N400), whose amplitude is modulated when a given word is not congruous in a context. A phonological rule of a mother tongue also creates predictions about phonologically legal forms in pseudowords. The predictive mechanism of the brain is supported by neurological studies: Bar et al. (2007) show the default mode or baseline of the brain areas (Raichle et al. 2001) overlap with the regions activated by tasks that recruit associative processing. This means associative activation equals the process of the brain's baseline state. Bar et al. (2007) propose that continuous generation of predictions is derived from associative processes and analogical mapping of the brain. This study proposes that language production and comprehension theories employ this predictive process of the brain in that prediction seems to affect speech errors as well as implicature recovery.

Objective: To identify early clinical predictors associated with dysphagia and delayed swallowing recovery in patients with traumatic brain injury (TBI).Methods: In this retrospective study, we enrolled adult TBI patients admitted to the rehabilitation unit of a tertiary medical center between June 2019 and June 2023. Data on baseline characteristics, neurological status, imaging findings, and rehabilitation-related variables were collected. Swallowing function was assessed using two indicators: (1) nasogastric (NG) tube retention and (2) the Functional Oral Intake Scale (FOIS) scores at 1, 4, and 12 weeks post-injury. Regression analyses were conducted to identify predictors associated with dysphagia and swallowing recovery.Results: A total of 160 patients were included. At 1 week post-injury, longer intensive care unit (ICU) stay, poor initial sitting balance and use of sedative medication in ICU were associated with NG tube retention. At 4 weeks, lower initial Rancho Los Amigos Scale (RLAS) scores, immobility-related complications, longer hospitalization, and temporal lobe hematomas were associated with persistent NG tube dependence. By 12 weeks, older age, delayed ability to follow commands, and poor initial sitting balance remained associated with NG tube retention. FOIS outcomes were also associated with older age, delayed time to follow commands, impaired initial sitting balance, prolonged ICU stay, temporal lobe hematomas, lower initial RLAS scores, immobility-related complications, prolonged endotracheal tube placement and extended hospital stays.Conclusion: Impaired cognitive status, poor physical function, immobility-related complications, and temporal lobe hematomas were key factors associated with dysphagia and delayed oral intake in individuals with TBI.

현대 사회의 급격한 고령화는 치매 환자의 폭발적인 증가를 초래하고 있으며, 이는 심각한 사회적 비용과 공중 보건 문제를 야기하고 있다. 특히 알츠하이머병(Alzheimer's Disease, AD)은 치매의 가장 주요한 원인으로, 완치 법이 부재한 현 상황에서는 조기 진단을 통한 진행 지연이 필수적이다. 본 연구는 뇌 MRI 영상 데이터를 활용하여 알츠하이머 질환을 정밀하게 예측할 수 있는 딥러닝 최적 모델을 제안하는 것을 목적으로 한다. 이를 위해 대표적인 합성곱 신경망(CNN) 모델인 VGG-19와 Inception ResNet V2를 선정하고, 의료 영상의 특성을 극대화할 수 있는 다양한 전처리 기법의 조합이 모델 성능에 미치는 영향을 심층 분석하였다. Kaggle의 알츠하이머 MRI 데이터 셋 6,400장을 활용한 실험 결과, VGG-19 모델에 ‘Overlap’ 전처리 기법을 단독으로 적용했을 때 검증 정확도 (Validation Accuracy) 0.98, F1 Score 0.98을 기록하여 가장 우수한 성능을 보였다.이는 임상적 진단 도구로서의 신뢰성을 높이는 데 더욱 효과적임을 시사한다.

The rapid aging of modern society has led to a sharp rise in dementia cases, creating significant social and public health burdens. Alzheimer’s Disease (AD), the leading cause of dementia, requires early diagnosis due to the absence of a cure. This study proposes an optimal deep learning model for precise AD prediction using brain MRI data. Utilizing 6,400 images from the Kaggle Alzheimer dataset, we comparatively evaluated VGG-19 and Inception-ResNet-V2 and analyzed the impact of various preprocessing strategies. Experimental results show that VGG-19 with the standalone ‘Overlap’ preprocessing technique achieved the best performance with a validation accuracy of 0.98 and an F1-score of 0.98, indicating strong potential as a reliable clinical diagnostic tool.

Objective : To determine whether the use of contrast enhancement (especially its extent) predicts malignant brain edema after intra-arterial thrombectomy (IAT) in patients with acute ischemic stroke. Methods : We reviewed the records of patients with acute ischemic stroke who underwent IAT for occlusion of the internal carotid artery or the middle cerebral artery between January 2012 and March 2015. To estimate the extent of contrast enhancement (CE), we used the contrast enhancement area ratio (CEAR)-i.e., the ratio of the CE to the area of the hemisphere, as noted on immediate non-enhanced brain computed tomography (NECT) post-IAT. Patients were categorized into two groups based on the CEAR values being either greater than or less than 0.2. Results : A total of 39 patients were included. Contrast enhancement was found in 26 patients (66.7%). In this subgroup, the CEAR was greater than 0.2 in 7 patients (18%) and less than 0.2 in the other 19 patients (48.7%). On univariate analysis, both CEAR ${\geq}0.2$ and the presence of subarachnoid hemorrhage were significantly associated with progression to malignant brain edema (p<0.001 and p=0.004), but on multivariate analysis, only CEAR ${\geq}0.2$ showed a statistically significant association (p=0.019). In the group with CEAR ${\geq}0.2$, the time to malignant brain edema was shorter (p=0.039) than in the group with CEAR <0.2. Clinical functional outcomes, based on the modified Rankin scale, were also significantly worse in patients with CEAR ${\geq}0.2$ (p=0.003) Conclusion : The extent of contrast enhancement as noted on NECT scans obtained immediately after IAT could be predictive of malignant brain edema and a poor clinical outcome.

This study aims to reveal the predictive processes in language comprehension―cognitive activities that occur unconsciously―by simply measuring response times. Following Tanenhaus et al. (1995), this study uses both auditory and visual stimuli, and uses parallel sentence structures to restrict contextual possibilities and encourage prediction. To compare the effects of prediction, half of the 50 participants were assigned to an ‘interference group’ to hinder natural prediction; the other half were assigned to a ‘prediction group.’ Both groups were instructed to mark the item they “heard” from a pair of options presented on paper. The results show that the prediction group completed the task, on average, 40 seconds faster than the interference group did. This difference happens because the interference group waited to actually hear the item before making a choice, while the prediction group marked the answer without waiting to hear it. Unlike previous studies on predictability, which often focus on words’ overall properties, this study targets lexical semantic categories. The stark difference in completion times between the two groups demonstrates how the brain’s predictive mechanisms can significantly accelerate language comprehension when not disrupted as in the case of the interference group.