Mapping the landscape of histomorphological cancer phenotypes using self-supervised learning on unannotated pathology slides.

Cancer diagnosis and management depend upon the extraction of complex information from microscopy images by pathologists, which requires time-consuming expert interpretation prone to human bias. Supervised deep learning approaches have proven powerful, but are inherently limited by the cost and quality of annotations used for training. Therefore, we present Histomorphological Phenotype Learning, a self-supervised methodology requiring no labels and operating via the automatic discovery of discriminatory features in image tiles. Tiles are grouped into morphologically similar clusters which constitute an atlas of histomorphological phenotypes (HP-Atlas), revealing trajectories from benign to malignant tissue via inflammatory and reactive phenotypes. These clusters have distinct features which can be identified using orthogonal methods, linking histologic, molecular and clinical phenotypes. Applied to lung cancer, we show that they align closely with patient survival, with histopathologically recognised tumor types and growth patterns, and with transcriptomic measures of immunophenotype. These properties are maintained in a multi-cancer study.

Inflammation in the tumor-adjacent lung as a predictor of clinical outcome in lung adenocarcinoma.

Approximately 30% of early-stage lung adenocarcinoma patients present with disease progression after successful surgical resection. Despite efforts of mapping the genetic landscape, there has been limited success in discovering predictive biomarkers of disease outcomes. Here we performed a systematic multi-omic assessment of 143 tumors and matched tumor-adjacent, histologically-normal lung tissue with long-term patient follow-up. Through histologic, mutational, and transcriptomic profiling of tumor and adjacent-normal tissue, we identified an inflammatory gene signature in tumor-adjacent tissue as the strongest clinical predictor of disease progression. Single-cell transcriptomic analysis demonstrated the progression-associated inflammatory signature was expressed in both immune and non-immune cells, and cell type-specific profiling in monocytes further improved outcome predictions. Additional analyses of tumor-adjacent transcriptomic data from The Cancer Genome Atlas validated the association of the inflammatory signature with worse outcomes across cancers. Collectively, our study suggests that molecular profiling of tumor-adjacent tissue can identify patients at high risk for disease progression.

MR spectroscopy frequency and phase correction using convolutional neural networks.

PURPOSE: To introduce a novel convolutional neural network (CNN)-based approach for frequency-and-phase correction (FPC) of MR spectroscopy (MRS) spectra to achieve fast and accurate FPC of single-voxel MEGA-PRESS MRS data. METHODS: Two neural networks (one for frequency and one for phase) were trained and validated using published simulated and in vivo MEGA-PRESS MRS dataset with wide-range artificial frequency and phase offsets applied. The CNN-based approach was subsequently tested and compared to the current deep learning solution: multilayer perceptrons (MLP). Furthermore, random noise was added to the original simulated dataset to further investigate the model performance at varied signal-to-noise ratio (SNR) levels (i.e., 10, 5, and 2.5). Additional frequency and phase offsets (i.e., small, moderate, large) were also applied to the in vivo dataset, and the CNN model was compared to the conventional approach SR and model-based SR implementation (mSR). RESULTS: The CNN model is more robust to noise compared to the MLP-based approach due to having smaller mean absolute errors in both frequency (0.01 ± 0.01 Hz at SNR = 10 and 0.01 ± 0.02 Hz at SNR = 2.5) and phase (0.12 ± 0.09° at SNR = 10 and -0.07 ± 0.44° at SNR = 2.5) offset prediction. Furthermore, better performance was demonstrated for FPC when compared to the MLP-based approach, and SR when applied to the in vivo dataset for both with and without additional offsets. CONCLUSION: A CNN-based approach provides a solution to the automated preprocessing of MRS data, and the experimental results demonstrate the quantitatively improved spectra quality compared to the state-of-the-art approach.

Mapping cognitive brain functions at scale.

A fundamental set of cognitive abilities enable humans to efficiently process goal-relevant information, suppress irrelevant distractions, maintain information in working memory, and act flexibly in different behavioral contexts. Yet, studies of human cognition and their underlying neural mechanisms usually evaluate these cognitive constructs in silos, instead of comprehensively in-tandem within the same individual. Here, we developed a scalable, mobile platform, "BrainE" (short for Brain Engagement), to rapidly assay several essential aspects of cognition simultaneous with wireless electroencephalography (EEG) recordings. Using BrainE, we rapidly assessed five aspects of cognition including (1) selective attention, (2) response inhibition, (3) working memory, (4) flanker interference and (5) emotion interference processing, in 102 healthy young adults. We evaluated stimulus encoding in all tasks using the EEG neural recordings, and isolated the cortical sources of the spectrotemporal EEG dynamics. Additionally, we used BrainE in a two-visit study in 24 young adults to investigate the reliability of the neuro-cognitive data as well as its plasticity to transcranial magnetic stimulation (TMS). We found that stimulus encoding on multiple cognitive tasks could be rapidly assessed, identifying common as well as distinct task processes in both sensory and cognitive control brain regions. Event related synchronization (ERS) in the theta (3-7 Hz) and alpha (8-12 Hz) frequencies as well as event related desynchronization (ERD) in the beta frequencies (13-30 Hz) were distinctly observed in each task. The observed ERS/ERD effects were overall anticorrelated. The two-visit study confirmed high test-retest reliability for both cognitive and neural data, and neural responses showed specific TMS protocol driven modulation. We also show that the global cognitive neural responses are sensitive to mental health symptom self-reports. This first study with the BrainE platform showcases its utility in studying neuro-cognitive dynamics in a rapid and scalable fashion.