Standardized CycleGAN training for unsupervised stain adaptation in invasive carcinoma classification for breast histopathology.

Purpose: Generalization is one of the main challenges of computational pathology. Slide preparation heterogeneity and the diversity of scanners lead to poor model performance when used on data from medical centers not seen during training. In order to achieve stain invariance in breast invasive carcinoma patch classification, we implement a stain translation strategy using cycleGANs for unsupervised image-to-image translation. Those models often suffer from a lack of proper metrics to monitor and stop the training at a particular point. We also introduce a method to solve this issue. Approach: We compare three CycleGAN-based approaches to a baseline classification model obtained without any stain invariance strategy. Two of the proposed approaches use CycleGAN's translations at inference or training to build stain-specific classification models. The last method uses them for stain data augmentation during training. This constrains the classification model to learn stain-invariant features. Regarding CycleGANs' training monitoring, we leverage Fréchet inception distance between generated and real samples and use it as a stopping criterion. We compare CycleGANs' models stopped using this criterion and models stopped at a fixed number of epochs. Results: Baseline metrics are set by training and testing the baseline classification model on a reference stain. We assessed performances using three medical centers with H&E and H&E&S staining. Every approach tested in this study improves baseline metrics without needing labels on target stains. The stain augmentation-based approach produced the best results on every stain. Each method's pros and cons are studied and discussed. Moreover, FID stopping criterion proves superiority to methods using a predefined number of training epoch and has the benefit of not requiring any visual inspection of CycleGAN results. Conclusion: We introduce a method to attain stain invariance for breast invasive carcinoma classification by leveraging CycleGAN's abilities to produce realistic translations between various stains. Moreover, we propose a systematical method for scheduling CycleGANs' trainings by using FID as a stopping criterion and prove its superiority to other methods. Finally, we give an insight on the minimal amount of data required for CycleGAN training in a digital histopathology setting.

Multicenter automatic detection of invasive carcinoma on breast whole slide images.

Breast cancer is one of the most prevalent cancers worldwide and pathologists are closely involved in establishing a diagnosis. Tools to assist in making a diagnosis are required to manage the increasing workload. In this context, artificial intelligence (AI) and deep-learning based tools may be used in daily pathology practice. However, it is challenging to develop fast and reliable algorithms that can be trusted by practitioners, whatever the medical center. We describe a patch-based algorithm that incorporates a convolutional neural network to detect and locate invasive carcinoma on breast whole-slide images. The network was trained on a dataset extracted from a reference acquisition center. We then performed a calibration step based on transfer learning to maintain the performance when translating on a new target acquisition center by using a limited amount of additional training data. Performance was evaluated using classical binary measures (accuracy, recall, precision) for both centers (referred to as "test reference dataset" and "test target dataset") and at two levels: patch and slide level. At patch level, accuracy, recall, and precision of the model on the reference and target test sets were 92.1% and 96.3%, 95% and 87.8%, and 73.9% and 70.6%, respectively. At slide level, accuracy, recall, and precision were 97.6% and 92.0%, 90.9% and 100%, and 100% and 70.8% for test sets 1 and 2, respectively. The high performance of the algorithm at both centers shows that the calibration process is efficient. This is performed using limited training data from the new target acquisition center and requires that the model is trained beforehand on a large database from a reference center. This methodology allows the implementation of AI diagnostic tools to help in routine pathology practice.

Large-Scale Functional Networks Identified from Resting-State EEG Using Spatial ICA.

Several methods have been applied to EEG or MEG signals to detect functional networks. In recent works using MEG/EEG and fMRI data, temporal ICA analysis has been used to extract spatial maps of resting-state networks with or without an atlas-based parcellation of the cortex. Since the links between the fMRI signal and the electromagnetic signals are not fully established, and to avoid any bias, we examined whether EEG alone was able to derive the spatial distribution and temporal characteristics of functional networks. To do so, we propose a two-step original method: 1) An individual multi-frequency data analysis including EEG-based source localisation and spatial independent component analysis, which allowed us to characterize the resting-state networks. 2) A group-level analysis involving a hierarchical clustering procedure to identify reproducible large-scale networks across the population. Compared with large-scale resting-state networks obtained with fMRI, the proposed EEG-based analysis revealed smaller independent networks thanks to the high temporal resolution of EEG, hence hierarchical organization of networks. The comparison showed a substantial overlap between EEG and fMRI networks in motor, premotor, sensory, frontal, and parietal areas. However, there were mismatches between EEG-based and fMRI-based networks in temporal areas, presumably resulting from a poor sensitivity of fMRI in these regions or artefacts in the EEG signals. The proposed method opens the way for studying the high temporal dynamics of networks at the source level thanks to the high temporal resolution of EEG. It would then become possible to study detailed measures of the dynamics of connectivity.

Expert and crowd-sourced validation of an individualized sleep spindle detection method employing complex demodulation and individualized normalization.

A spindle detection method was developed that: (1) extracts the signal of interest (i.e., spindle-related phasic changes in sigma) relative to ongoing "background" sigma activity using complex demodulation, (2) accounts for variations of spindle characteristics across the night, scalp derivations and between individuals, and (3) employs a minimum number of sometimes arbitrary, user-defined parameters. Complex demodulation was used to extract instantaneous power in the spindle band. To account for intra- and inter-individual differences, the signal was z-score transformed using a 60 s sliding window, per channel, over the course of the recording. Spindle events were detected with a z-score threshold corresponding to a low probability (e.g., 99th percentile). Spindle characteristics, such as amplitude, duration and oscillatory frequency, were derived for each individual spindle following detection, which permits spindles to be subsequently and flexibly categorized as slow or fast spindles from a single detection pass. Spindles were automatically detected in 15 young healthy subjects. Two experts manually identified spindles from C3 during Stage 2 sleep, from each recording; one employing conventional guidelines, and the other, identifying spindles with the aid of a sigma (11-16 Hz) filtered channel. These spindles were then compared between raters and to the automated detection to identify the presence of true positives, true negatives, false positives and false negatives. This method of automated spindle detection resolves or avoids many of the limitations that complicate automated spindle detection, and performs well compared to a group of non-experts, and importantly, has good external validity with respect to the extant literature in terms of the characteristics of automatically detected spindles.