Dual resolution deep learning network with self-attention mechanism for classification and localisation of colorectal cancer in histopathological images

Abstract

AimsMicroscopic examination is a basic diagnostic technology for colorectal cancer (CRC), but it is very laborious. We developed a dual resolution deep learning network with self-attention mechanism (DRSANet) which combines context and details for CRC binary classification and localisation in whole slide images (WSIs), and as a computer-aided diagnosis (CAD) to improve the sensitivity and specificity of doctors’ diagnosis.MethodsRepresentative regions of interest (ROI) of each tissue type were manually delineated in WSIs by pathologists. Based on the same coordinates of centre position, patches were extracted at different magnification levels from the ROI. Specifically, patches from low magnification level contain contextual information, while from high magnification level provide important details. A dual-inputs network was designed to learn context and details simultaneously, and self-attention mechanism was used to selectively learn different positions in the images to enhance the performance.ResultsIn classification task, DRSANet outperformed the benchmark networks which only depended on the high magnification patches on two test set. Furthermore, in localisation task, DRSANet demonstrated a better localisation capability of tumour area in WSI with less areas of misidentification.ConclusionsWe compared DRSANet with benchmark networks which only use the patches from high magnification level. Experimental results reveal that the performance of DRSANet is better than the benchmark networks. Both context and details should be considered in deep learning method.

Deep learning for subtypes identification of pure seminoma of the testis

“…We hypothesize that consideration of seminoma subtypes during development of treatment strategy will potentially improve its clinical management and implementation of developed model will improve diagnosis accuracy and exclude potential errors. Our model achieved the highest validation accuracy with 0.933, which is comparable with accuracy of other DL binary classifiers of histopathological images [20].…”

Artificial intelligence in digital pathology: a systematic review and meta-analysis of diagnostic test accuracy

“…Subgroups of gastrointestinal pathology, breast pathology and urological pathology studies were examined in more detail, as these were the largest subsets of studies identified (see Table 8 and Supplementary Materials ). The gastrointestinal subgroup demonstrated high mean sensitivity and specificity and included AI models for colorectal cancer 28 – 30 , 32 , 34 , 40 , gastric cancer 28 , 31 , 33 , 37 – 39 , 85 and gastritis 35 . The breast subgroup included only AI models for breast cancer applications, with Hameed et al and Wang et al demonstrating particularly high sensitivity (98%, 91% respectively) and specificity (93%, 96% respectively) 42 , 45 .…”

Multi-head self-attention mechanism enabled individualized hemoglobin prediction and treatment recommendation systems in anemia management for hemodialysis patients

“…Lee et al used SAM to process irregular multivariate time-series data in the EHR to predict in-hospital mortality, length of stay, and phenotyping [ 24 ]. Xu Y et al used SAM to selectively learn different positions in pathological slide images to improve the performance of colorectal cancer diagnosis [ 25 ]. Wang et al applied SAM to the lesion segmentation network on chest CT images to diagnose COVID-19 [ 26 ].…”