Development of a multiple convolutional neural network-facilitated diagnostic screening program for immunofluorescence images of IgA nephropathy and idiopathic membranous nephropathy.

Background: Immunoglobulin A nephropathy (IgAN) and idiopathic membranous nephropathy (IMN) are the most common glomerular diseases. Immunofluorescence (IF) tests of renal tissues are crucial for the diagnosis. We developed a multiple convolutional neural network (CNN)-facilitated diagnostic program to assist the IF diagnosis of IgAN and IMN. Methods: The diagnostic program consisted of four parts: a CNN trained as a glomeruli detection module, an IF intensity comparator, dual-CNN (D-CNN) trained as a deposition appearance and location classifier and a post-processing module. A total of 1573 glomerular IF images from 1009 patients with glomerular diseases were used for the training and validation of the diagnostic program. A total of 1610 images of 426 patients from different hospitals were used as test datasets. The performance of the diagnostic program was compared with nephropathologists. Results: In >90% of the tested images, the glomerulus location module achieved an intersection over union >0.8. The accuracy of the D-CNN in recognizing irregular granular mesangial deposition and fine granular deposition along the glomerular basement membrane was 96.1% and 93.3%, respectively. As for the diagnostic program, the accuracy, sensitivity and specificity of diagnosing suspected IgAN were 97.6%, 94.4% and 96.0%, respectively. The accuracy, sensitivity and specificity of diagnosing suspected IMN were 91.7%, 88.9% and 95.8%, respectively. The corresponding areas under the curve (AUCs) were 0.983 and 0.935. When tested with images from the outside hospital, the diagnostic program showed stable performance. The AUCs for diagnosing suspected IgAN and IMN were 0.972 and 0.948, respectively. Compared with inexperienced nephropathologists, the program showed better performance. Conclusion: The proposed diagnostic program could assist the IF diagnosis of IgAN and IMN.

Sparse and Hierarchical Transformer for Survival Analysis on Whole Slide Images.

The Transformer-based methods provide a good opportunity for modeling the global context of gigapixel whole slide image (WSI), however, there are still two main problems in applying Transformer to WSI-based survival analysis task. First, the training data for survival analysis is limited, which makes the model prone to overfitting. This problem is even worse for Transformer-based models which require large-scale data to train. Second, WSI is of extremely high resolution (up to 150,000 x 150,000 pixels) and is typically organized as a multi-resolution pyramid. Vanilla Transformer cannot model the hierarchical structure of WSI (such as patch cluster-level relationships), which makes it incapable of learning hierarchical WSI representation. To address these problems, in this paper, we propose a novel Sparse and Hierarchical Transformer (SH-Transformer) for survival analysis. Specifically, we introduce sparse self-attention to alleviate the overfitting problem, and propose a hierarchical Transformer structure to learn the hierarchical WSI representation. Experimental results based on three WSI datasets show that the proposed framework outperforms the state-of-the-art methods.

TransSurv: Transformer-based Survival Analysis Model Integrating Histopathological Images and Genomic Data for Colorectal Cancer.

Survival analysis is a significant study in cancer prognosis, and the multi-modal data, including histopathological images, genomic data, and clinical information, provides unprecedented opportunities for its development. However, because of the high dimensionality and the heterogeneity of histopathological images and genomic data, acquiring effective predictive characters from these multi-modal data has always been a challenge for survival analysis. In this study, we propose a transformer-based survival analysis model (TransSurv) for colorectal cancer that can effectively integrate intra-modality and inter-modality features of histopathological images, genomic data, and clinical information. Specifically, to integrate the intra-modality relationship of image patches, we develop a multi-scale histopathological features fusion transformer (MS-Trans). Furthermore, we provide a cross-modal fusion transformer based on cross attention for multi-scale pathological representation and multi-omics representation, which includes RNA-seq expression and copy number alteration (CNA). At the output layer of the TransSurv, we adopt the Cox layer to integrate multi-modal fusion representation with clinical information for end-to-end survival analysis. The experimental results on the Cancer Genome Atlas (TCGA) colorectal cancer cohort demonstrate that the proposed TransSurv outperforms the existing methods and improves the prognosis prediction of colorectal cancer.

Nuclei-Guided Network for Breast Cancer Grading in HE-Stained Pathological Images.

Breast cancer grading methods based on hematoxylin-eosin (HE) stained pathological images can be summarized into two categories. The first category is to directly extract the pathological image features for breast cancer grading. However, unlike the coarse-grained problem of breast cancer classification, breast cancer grading is a fine-grained classification problem, so general methods cannot achieve satisfactory results. The second category is to apply the three evaluation criteria of the Nottingham Grading System (NGS) separately, and then integrate the results of the three criteria to obtain the final grading result. However, NGS is only a semiquantitative evaluation method, and there may be far more image features related to breast cancer grading. In this paper, we proposed a Nuclei-Guided Network (NGNet) for breast invasive ductal carcinoma (IDC) grading in pathological images. The proposed nuclei-guided attention module plays the role of nucleus attention, so as to learn more nuclei-related feature representations for breast IDC grading. In addition, the proposed nuclei-guided fusion module in the fusion process of different branches can further enable the network to focus on learning nuclei-related features. Overall, under the guidance of nuclei-related features, the entire NGNet can learn more fine-grained features for breast IDC grading. The experimental results show that the performance of the proposed method is better than that of state-of-the-art method. In addition, we released a well-labeled dataset with 3644 pathological images for breast IDC grading. This dataset is currently the largest publicly available breast IDC grading dataset and can serve as a benchmark to facilitate a broader study of breast IDC grading.

Richer fusion network for breast cancer classification based on multimodal data.

BACKGROUND: Deep learning algorithms significantly improve the accuracy of pathological image classification, but the accuracy of breast cancer classification using only single-mode pathological images still cannot meet the needs of clinical practice. Inspired by the real scenario of pathologists reading pathological images for diagnosis, we integrate pathological images and structured data extracted from clinical electronic medical record (EMR) to further improve the accuracy of breast cancer classification. METHODS: In this paper, we propose a new richer fusion network for the classification of benign and malignant breast cancer based on multimodal data. To make pathological image can be integrated more sufficient with structured EMR data, we proposed a method to extract richer multilevel feature representation of the pathological image from multiple convolutional layers. Meanwhile, to minimize the information loss for each modality before data fusion, we use the denoising autoencoder as a way to increase the low-dimensional structured EMR data to high-dimensional, instead of reducing the high-dimensional image data to low-dimensional before data fusion. In addition, denoising autoencoder naturally generalizes our method to make the accurate prediction with partially missing structured EMR data. RESULTS: The experimental results show that the proposed method is superior to the most advanced method in terms of the average classification accuracy (92.9%). In addition, we have released a dataset containing structured data from 185 patients that were extracted from EMR and 3764 paired pathological images of breast cancer, which can be publicly downloaded from http://ear.ict.ac.cn/?page_id=1663 . CONCLUSIONS: We utilized a new richer fusion network to integrate highly heterogeneous data to leverage the structured EMR data to improve the accuracy of pathological image classification. Therefore, the application of automatic breast cancer classification algorithms in clinical practice becomes possible. Due to the generality of the proposed fusion method, it can be straightforwardly extended to the fusion of other structured data and unstructured data.

A Parallel Cerebrovascular Segmentation Algorithm Based on Focused Multi-Gaussians Model and Heterogeneous Markov Random Field.

A complete and detailed cerebrovascular image segmented from time-of-flight magnetic resonance angiography (TOF-MRA) data is essential for the diagnosis and therapy of the cerebrovascular diseases. In recent years, three-dimensional cerebrovascular segmentation algorithms based on statistical models have been widely used, but the existed methods always perform poorly on stenotic vessels and are not robust enough. In this paper, we propose a parallel cerebrovascular segmentation algorithm based on focused multi-Gaussians model and heterogeneous Markov random field. Specifically, we present a focused multi-Gaussians (FMG) model with local fitting region to model the vascular tissue more accurately and introduce the chaotic oscillation particle swarm optimization (CO-PSO) algorithm to improve the global optimization capability in the parameter estimation. Furthermore, we design a heterogeneous Markov Random Field (MRF) in the three-dimensional neighborhood system to incorporate precise local character of image. Finally, the algorithm has been performed parallel optimization based on GPUs and obtain about 60 times speedup compared to serial execution. The experiments show that the proposed algorithm can produce more detailed segmentation result in shorter time and performs well on the stenotic vessels robustly.

Super-resolution nanoscopy by coherent control on nanoparticle emission.

Super-resolution nanoscopy based on wide-field microscopic imaging provided high efficiency but limited resolution. Here, we demonstrate a general strategy to push its resolution down to ~50 nm, which is close to the range of single molecular localization microscopy, without sacrificing the wide-field imaging advantage. It is done by actively and simultaneously modulating the characteristic emission of each individual emitter at high density. This method is based on the principle of excited state coherent control on single-particle two-photon fluorescence. In addition, the modulation efficiently suppresses the noise for imaging. The capability of the method is verified both in simulation and in experiments on ZnCdS quantum dot-labeled films and COS7 cells. The principle of coherent control is generally applicable to single-multiphoton imaging and various probes.