BACKGROUND:The role of cervical lymph node dissection in esophageal squamous cell carcinoma (ESCC) surgery remains controversial. This study aims to explore a machine learning (ML) approach that integrates sonographic and clinical findings for the noninvasive evaluation of cervical lymph node involvement in patients with ESCC. MATERIALS AND METHODS:The dataset contained 887 ESCC patients who underwent surgery and subsequent pathological examination of their cervical lymph nodes. Selected ML models were established to predict metastasis using patient characteristics and ultrasound evaluations. The models were tested through fivefold cross-validation and benchmarked against a baseline nomogram. The importance of each predictor variable was quantified by the permutation score. RESULTS:Of the patients, 32.1% had confirmed cervical nodal metastasis. The random forest model exhibited a mean accuracy of 0.68 (95% confidence interval: 0.65-0.71), area under the curve of 0.72 (0.71-0.74), and F1 score of 0.56 (0.54-0.58), comparable to other models ( p > 0.1). All ML models showed superior predictive power over the baseline nomogram ( p < 0.05). The most critical predictor was the maximum cervical nodal diameter from ultrasound. Models using comprehensive baseline features surpassed those with sonographic data alone ( p < 0.001), while intraoperative pathology did not improve predictions. CONCLUSIONS:This study highlights the diagnostic value of ultrasonography for cervical nodal metastasis in ESCC and proposes an ML-based noninvasive method to inform decisions on lymph node dissection. The predictive model may enhance surgical planning and enable personalized treatment strategies.
Sleep decoding is key to revealing sleep architecture and its links to health, yet prevailing deep-learning models rely on supervised, task-specific designs and dual encoders that isolate time-domain and frequency-domain information, limiting generalizability and scalability. We introduce SleepGPT for sleep decoding, a time-frequency foundation model based on generative pretrained transformer, developed using multi-pretext pretraining strategy on 86,335 hours of polysomnography (PSG) from 8,377 subjects. SleepGPT includes a channel-adaptive mechanism for variable channel configurations and a unified time-frequency fusion module that enables deep cross-domain interaction. Evaluations across diverse PSG datasets demonstrate that SleepGPT sets a new benchmark for sleep decoding tasks, achieving superior performance in sleep staging, sleep-related pathology classification, sleep data generation, and sleep spindle detection. Moreover, it reveals channel- and stage-specific physiological patterns underlying sleep decoding. In sum, SleepGPT is an all-in-one method with exceptional generalizability and scalability, offering transformative potential in addressing sleep decoding challenges.
Humans excel at learning from sparse experience by leveraging internal models of the world to infer the values of options they have never sampled, i.e., model-based learning. However, how the brain supports such learning remains largely unknown. Here, we recorded intracranial electrophysiology (iEEG) from 34 epilepsy patients performing a reinforcement-learning task that required using the task structure to infer the values of unvisited (non-local) paths. Hippocampal ripples were associated with prioritized non-local learning. After each outcome, ripple events carried information about which indirect experience was most valuable to update, with longer ripples showing stronger priority signals. These events coincided with stronger cortical reactivation of high-priority than low-priority paths. Importantly, lateral frontopolar cortex activity precisely synchronized with hippocampal ripples in this post-reward time window, which predicted more effective use of task structure and more accurate non-local value learning. Together, our findings suggest that ripple-centered hippocampal-prefrontal coordination supports efficient model-based learning.
Dear Editor, The developing cerebral cortex requires precise metabolic regulation to support neurogenesis and circuit formation (Belanger et al., 2011; Namba et al., 2021). Glutamine syn thetase (GS), which catalyzes glutamate-to-glutamine conversion, sustains neurotransmitter recycling and nitro gen homeostasis (Tani et al., 2014). Human GLUL muta tions cause lethal neurodevelopmental disorders (H?berle et al., 2005), and GS deletion in mice leads to postnatal death (He et al., 2010), underscoring its essential role. While adult cortical GS deficiency triggers neurodegener ation (Zhou et al., 2019), its function in early cortical devel opment remains elusive, despite evidence linking astrocytic metabolism to circuit maturation (Chung et al., 2015; Pekny et al., 2016).
Three-dimensional (3D) optical imaging of intact tissues with high spatiotemporal resolution is a highly desirable goal, as it offers visualization of biological structures and processes in their physiological context. Optical sectioning structured illumination microscopy (OS-SIM) stands out among various 3D imaging techniques for its comprehensive superiority, as it offers high spatial resolution, rapid imaging speed, and low phototoxicity. However, under the challenging conditions of dominant background and increased fluorescence scattering in thick tissues, which leads to a decrease in modulation contrast, OS-SIM faces the problem of low signal-to-noise ratio (SNR) and limited imaging depth. We have developed a high SNR OS-SIM strategy for 3D imaging of biological samples ranging from millimeter-thick tissues to cellular dynamics. By enhancing the SNR by ∼10 dB with rapid processing, our method enables imaging of tissue-cleared sample with thickness >2 mm and low light dose for live-cell recording, while also doubling the imaging penetration depth. With our OS-SIM system, we achieve high-quality 3D images of various biological samples, including mouse brains, Drosophila clock neurons, organoids, and live cells. We anticipate that this approach will render OS-SIM a powerful technique for the research of cellular organelles or thick tissues in 3D morphology.
