Recent application of artificial intelligence on histopathologic image-based prediction of gene mutation in solid cancers.

PURPOSE: Evaluation of genetic mutations in cancers is important because distinct mutational profiles help determine individualized drug therapy. However, molecular analyses are not routinely performed in all cancers because they are expensive, time-consuming and not universally available. Artificial intelligence (AI) has shown the potential to determine a wide range of genetic mutations on histologic image analysis. Here, we assessed the status of mutation prediction AI models on histologic images by a systematic review. METHODS: A literature search using the MEDLINE, Embase and Cochrane databases was conducted in August 2021. The articles were shortlisted by titles and abstracts. After a full-text review, publication trends, study characteristic analysis and comparison of performance metrics were performed. RESULTS: Twenty-four studies were found mostly from developed countries, and their number is increasing. The major targets were gastrointestinal, genitourinary, gynecological, lung and head and neck cancers. Most studies used the Cancer Genome Atlas, with a few using an in-house dataset. The area under the curve of some of the cancer driver gene mutations in particular organs was satisfactory, such as 0.92 of BRAF in thyroid cancers and 0.79 of EGFR in lung cancers, whereas the average of all gene mutations was 0.64, which is still suboptimal. CONCLUSION: AI has the potential to predict gene mutations on histologic images with appropriate caution. Further validation with larger datasets is still required before AI models can be used in clinical practice to predict gene mutations.

Deep learning captures selective features for discrimination of microsatellite instability from pathologic tissue slides of gastric cancer.

Microsatellite instability (MSI) status is an important prognostic marker for various cancers. Furthermore, because immune checkpoint inhibitors are much more effective in tumors with high level of MSI (MSI-H), MSI status is routinely tested in multiple cancer types. Therefore, many studies have tested the feasibility of deep learning (DL)-based prediction of MSI status from hematoxylin and eosin (H&E)-stained tissue slides. In the present study, we attempted a fully automated classification of MSI status in gastric cancer (GC) tissue slides. For frozen and formalin-fixed paraffin-embedded (FFPE) GC tissues from The Cancer Genome Atlas (TCGA), the areas under the curves (AUCs) for the receiver operating characteristic (ROC) curves were 0.893 and 0.902, respectively. The classifier trained with the TCGA FFPE tissues performed well on an external validation Asian FFPE cohort, with an AUC of 0.874. However, the DL-based classifier seems incompatible with cancers from different organs because morphologic features of MSI-H tissues are different. Analysis of histomorphologic features of MSI-H GC tissues suggested that MSI-H GC could largely be divided into two groups: intestinal type tumors with moderate to poor differentiation and diffuse type mucinous tumors. However, the recognizable morphologic features cannot completely explain the good performance of the DL-based classifier. These results indicate that DL could automatically learn the optimal features for discrimination of MSI status in GC tissue slides. This study demonstrated the potential of a DL-based MSI classifier as a screening tool for definitive cases.

Group 1 metabotropic glutamate receptor 5 is involved in synaptically-induced Ca2+-spikes and cell death in cultured rat hippocampal neurons.

Group 1 metabotropic glutamate receptors (mGluRs) can positively affect postsynaptic neuronal excitability and epileptogenesis. The objective of the present study was to determine whether group 1 mGluRs might be involved in synaptically-induced intracellular free Ca2+ concentration ([Ca2+]i) spikes and neuronal cell death induced by 0.1 mM Mg2+ and 10 µM glycine in cultured rat hippocampal neurons from embryonic day 17 fetal Sprague-Dawley rats using imaging methods for Ca2+ and 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide assays for cell survival. Reduction of extracellular Mg2+ concentration ([Mg2+]o) to 0.1 mM induced repetitive [Ca2+]i spikes within 30 sec at day 11.5. The mGluR5 antagonist 6-Methyl-2-(phenylethynyl) pyridine (MPEP) almost completely inhibited the [Ca2+]i spikes, but the mGluR1 antagonist LY367385 did not. The group 1 mGluRs agonist, 3,5-dihydroxyphenylglycine (DHPG), significantly increased the [Ca2+]i spikes. The phospholipase C inhibitor U73122 significantly inhibited the [Ca2+]i spikes in the absence or presence of DHPG. The IP3 receptor antagonist 2-aminoethoxydiphenyl borate or the ryanodine receptor antagonist 8-(diethylamino)octyl 3,4,5-trimethoxybenzoate also significantly inhibited the [Ca2+]i spikes in the absence or presence of DHPG. The TRPC channel inhibitors SKF96365 and flufenamic acid significantly inhibited the [Ca2+]i spikes in the absence or presence of DHPG. The mGluR5 antagonist MPEP significantly increased the neuronal cell survival, but mGluR1 antagonist LY367385 did not. These results suggest a possibility that mGluR5 is involved in synaptically-induced [Ca2+]i spikes and neuronal cell death in cultured rat hippocampal neurons by releasing Ca2+ from IP3 and ryanodine-sensitive intracellular stores and activating TRPC channels.

Feasibility of deep learning-based fully automated classification of microsatellite instability in tissue slides of colorectal cancer.

High levels of microsatellite instability (MSI-H) occurs in about 15% of sporadic colorectal cancer (CRC) and is an important predictive marker for response to immune checkpoint inhibitors. To test the feasibility of a deep learning (DL)-based classifier as a screening tool for MSI status, we built a fully automated DL-based MSI classifier using pathology whole-slide images (WSIs) of CRCs. On small image patches of The Cancer Genome Atlas (TCGA) CRC WSI dataset, tissue/non-tissue, normal/tumor and MSS/MSI-H classifiers were applied sequentially for the fully automated prediction of the MSI status. The classifiers were also tested on an independent cohort. Furthermore, to test how the expansion of the training data affects the performance of the DL-based classifier, additional classifier trained on both TCGA and external datasets was tested. The areas under the receiver operating characteristic curves were 0.892 and 0.972 for the TCGA and external datasets, respectively, by a classifier trained on both datasets. The performance of the DL-based classifier was much better than that of previously reported histomorphology-based methods. We speculated that about 40% of CRC slides could be screened for MSI status without molecular testing by the DL-based classifier. These results demonstrated that the DL-based method has potential as a screening tool to discriminate molecular alteration in tissue slides.

Feasibility of fully automated classification of whole slide images based on deep learning.

Although microscopic analysis of tissue slides has been the basis for disease diagnosis for decades, intra- and inter-observer variabilities remain issues to be resolved. The recent introduction of digital scanners has allowed for using deep learning in the analysis of tissue images because many whole slide images (WSIs) are accessible to researchers. In the present study, we investigated the possibility of a deep learning-based, fully automated, computer-aided diagnosis system with WSIs from a stomach adenocarcinoma dataset. Three different convolutional neural network architectures were tested to determine the better architecture for tissue classifier. Each network was trained to classify small tissue patches into normal or tumor. Based on the patch-level classification, tumor probability heatmaps can be overlaid on tissue images. We observed three different tissue patterns, including clear normal, clear tumor and ambiguous cases. We suggest that longer inspection time can be assigned to ambiguous cases compared to clear normal cases, increasing the accuracy and efficiency of histopathologic diagnosis by pre-evaluating the status of the WSIs. When the classifier was tested with completely different WSI dataset, the performance was not optimal because of the different tissue preparation quality. By including a small amount of data from the new dataset for training, the performance for the new dataset was much enhanced. These results indicated that WSI dataset should include tissues prepared from many different preparation conditions to construct a generalized tissue classifier. Thus, multi-national/multi-center dataset should be built for the application of deep learning in the real world medical practice.

The Neuroprotective Effect of Hericium erinaceus Extracts in Mouse Hippocampus after Pilocarpine-Induced Status Epilepticus.

Hericium erinaceus (HE), a culinary-medicinal mushroom, has shown therapeutic potential in many brain diseases. However, the role of HE in status epilepticus (SE)-mediated neuronal death and its underlying mechanisms remain unclear. We investigated the neuroprotective effects of HE using a pilocarpine-induced SE model. Male C57BL/6 mice received crude extracts of HE (60 mg/kg, 120 mg/kg, or 300 mg/kg, p.o.) for 21 d from 14 d before SE to 6 d after SE. At 7 d after SE, cresyl violet and immunohistochemistry of neuronal nuclei revealed improved hippocampal neuronal survival in animals treated with 60 mg/kg and 120 mg/kg of HE, whereas those treated with 300 mg/kg of HE showed similar neuronal death to that of vehicle-treated controls. While seizure-induced reactive gliosis, assessed by immunohistochemistry, was not altered by HE, the number of hippocampal cyclooxygenase 2 (COX2)-expressing cells was significantly reduced by 60 and 120 mg/kg of HE. Triple immunohistochemistry demonstrated no overlap of COX2 labeling with Ox42, in addition to a decrease in COX2/GFAP-co-immunoreactivity in the group treated with 60 mg/kg HE, suggesting that the reduction of COX2 by HE promotes neuroprotection after SE. Our findings highlight the potential application of HE for preventing neuronal death after seizures.

Dual deep neural network-based classifiers to detect experimental seizures.

Manually reviewing electroencephalograms (EEGs) is labor-intensive and demands automated seizure detection systems. To construct an efficient and robust event detector for experimental seizures from continuous EEG monitoring, we combined spectral analysis and deep neural networks. A deep neural network was trained to discriminate periodograms of 5-sec EEG segments from annotated convulsive seizures and the pre- and post-EEG segments. To use the entire EEG for training, a second network was trained with non-seizure EEGs that were misclassified as seizures by the first network. By sequentially applying the dual deep neural networks and simple pre- and post-processing, our autodetector identified all seizure events in 4,272 h of test EEG traces, with only 6 false positive events, corresponding to 100% sensitivity and 98% positive predictive value. Moreover, with pre-processing to reduce the computational burden, scanning and classifying 8,977 h of training and test EEG datasets took only 2.28 h with a personal computer. These results demonstrate that combining a basic feature extractor with dual deep neural networks and rule-based pre- and post-processing can detect convulsive seizures with great accuracy and low computational burden, highlighting the feasibility of our automated seizure detection algorithm.

The level of nitric oxide regulates lipocalin-2 expression under inflammatory condition in RINm5F beta-cells.

We previously reported that proinflammatory cytokines (interleukin-1ß and interferon-γ) induced the expression of lipocalin-2 (LCN-2) together with inducible nitric oxide synthase (iNOS) in RINm5F beta-cells. Therefore, we examined the effect of nitric oxide (NO) on LCN-2 expression in cytokines-treated RINm5F beta-cells. Additionally, we observed the effect of LCN-2 on cell viability. First, we found the existence of LCN-2 receptor and the internalization of exogenous recombinant LCN-2 peptide in RINm5F and INS-1 beta-cells. Next, the effects of NO on LCN-2 expression were evaluated. Aminoguanidine, an iNOS inhibitor and iNOS gene silencing significantly inhibited cytokines-induced LCN-2 expression while sodium nitroprusside (SNP), an NO donor potentiated it. Luciferase reporter assay showed that transcription factor NF-κB was not involved in LCN-2 expression. Both LCN-2 mRNA and protein stability assays were conducted. SNP did not affect LCN-2 mRNA stability, however, it significantly reduced LCN-2 protein degradation. The LCN-2 protein degradation was significantly attenuated by MG132, a proteasome inhibitor. Finally, the effect of LCN-2 on cell viability was evaluated. LCN-2 peptide treatment and LCN-2 overexpression significantly reduced cell viability. FACS analysis showed that LCN-2 induced the apoptosis of the cells. Collectively, NO level affects LCN-2 expression via regulation of LCN-2 protein stability under inflammatory condition and LCN-2 may reduce beta-cell viability by promoting apoptosis.

GluN2B-containing N-methyl-D-aspartate receptors compensate for the inhibitory control of synaptic plasticity during the early critical period in the rat visual cortex.

In the visual cortex, synaptic plasticity is very high during the early developmental stage known as the critical period and declines with development after the critical period. Changes in the properties of N-methyl-D-aspartate receptor (NMDAR) and γ-aminobutyric acid type A receptor (GABAA R) have been suggested to underlie the changes in the characteristics of plasticity. However, it is largely unknown how the changes in the two receptors interact to regulate synaptic plasticity. The present study investigates the changes in the properties of NMDAR and GABAA R from 3 to 5 weeks of age in layer 2/3 pyramidal neurons of the rat visual cortex. The impact of these changes on the characteristics of long-term potentiation (LTP) is also investigated. The amplitude and decay time constant of GABAA R-mediated currents increased during this period. However, the decay time constant of NMDAR-mediated currents decreased as a result of the decrease in the proportion of the GluN2B subunit-mediated component. Induction of NMDAR-dependent LTP at 3 weeks depended on the GluN2B subunit, but LTP at 5 weeks did not. Enhancement of GABAA R-mediated inhibition suppressed the induction of LTP only at 5 weeks. However, partial inhibition of the GluN2B subunit with a low concentration of ifenprodil allowed the GABAA R-mediated suppression of LTP at 3 weeks. These results suggest that changes in the properties of NMDAR- and GABAA R-mediated synaptic transmission interact to determine the characteristics of synaptic plasticity during the critical period in the visual cortex.

Enhancement of GluN2B Subunit-Containing NMDA Receptor Underlies Serotonergic Regulation of Long-Term Potentiation after Critical Period in the Rat Visual Cortex.

Serotonin [5-hydroxytryptamine (5-HT)] regulates synaptic plasticity in the visual cortex. Although the effects of 5-HT on plasticity showed huge diversity depending on the ages of animals and species, it has been unclear how 5-HT can show such diverse effects. In the rat visual cortex, 5-HT suppressed long-term potentiation (LTP) at 5 weeks but enhanced LTP at 8 weeks. We speculated that this difference may originate from differential regulation of neurotransmission by 5-HT between the age groups. Thus, we investigated the effects of 5-HT on apha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptor (AMPAR)-, γ-aminobutyric acid receptor type A (GABAAR)-, and N-methyl-D-aspartic acid receptor (NMDAR)-mediated neurotransmissions and their involvement in the differential regulation of plasticity between 5 and 8 weeks. AMPAR-mediated currents were not affected by 5-HT at both 5 and 8 weeks. GABAAR-mediated currents were enhanced by 5-HT at both age groups. However, 5-HT enhanced NMDAR-mediated currents only at 8 weeks. The enhancement of NMDAR-mediated currents appeared to be mediated by the enhanced function of GluN2B subunit-containing NMDAR. The enhanced GABAAR- and NMDAR-mediated neurotransmissions were responsible for the suppression of LTP at 5 weeks and the facilitation of LTP at 8 weeks, respectively. These results indicate that the effects of 5-HT on neurotransmission change with development, and the changes may underlie the differential regulation of synaptic plasticity between different age groups. Thus, the developmental changes in 5-HT function should be carefully considered while investigating the 5-HT-mediated metaplastic control of the cortical network.

Phasic and Tonic Inhibition are Maintained Respectively by CaMKII and PKA in the Rat Visual Cortex.

Phasic and tonic γ-aminobutyric acidA (GABAA) receptor-mediated inhibition critically regulate neuronal information processing. As these two inhibitory modalities have distinctive features in their receptor composition, subcellular localization of receptors, and the timing of receptor activation, it has been thought that they might exert distinct roles, if not completely separable, in the regulation of neuronal function. Inhibition should be maintained and regulated depending on changes in network activity, since maintenance of excitation-inhibition balance is essential for proper functioning of the nervous system. In the present study, we investigated how phasic and tonic inhibition are maintained and regulated by different signaling cascades. Inhibitory postsynaptic currents were measured as either electrically evoked events or spontaneous events to investigate regulation of phasic inhibition in layer 2/3 pyramidal neurons of the rat visual cortex. Tonic inhibition was assessed as changes in holding currents by the application of the GABAA receptor blocker bicuculline. Basal tone of phasic inhibition was maintained by intracellular Ca(2+) and Ca(2+)/calmodulin-dependent protein kinase II (CaMKII). However, maintenance of tonic inhibition relied on protein kinase A activity. Depolarization of membrane potential (5 min of 0 mV holding) potentiated phasic inhibition via Ca(2+) and CaMKII but tonic inhibition was not affected. Thus, phasic and tonic inhibition seem to be independently maintained and regulated by different signaling cascades in the same cell. These results suggest that neuromodulatory signals might differentially regulate phasic and tonic inhibition in response to changes in brain states.

Cholinergic induction of input-specific late-phase LTP via localized Ca2+ release in the visual cortex.

Acetylcholine facilitates long-term potentiation (LTP) and long-term depression (LTD), substrates of learning, memory, and sensory processing, in which acetylcholine also plays a crucial role. Ca(2+) ions serve as a canonical regulator of LTP/LTD but little is known about the effect of acetylcholine on intracellular Ca(2+) dynamics. Here, we investigated dendritic Ca(2+) dynamics evoked by synaptic stimulation and the resulting LTP/LTD in layer 2/3 pyramidal neurons of the rat visual cortex. Under muscarinic stimulation, single-shock electrical stimulation (SES) inducing ∼20 mV EPSP, applied via a glass electrode located ∼10 µm from the basal dendrite, evoked NMDA receptor-dependent fast Ca(2+) transients and the subsequent Ca(2+) release from the inositol 1,4,5-trisphosphate (IP(3))-sensitive stores. These secondary dendritic Ca(2+) transients were highly localized within 10 µm from the center (SD = 5.0 µm). The dendritic release of Ca(2+) was a prerequisite for input-specific muscarinic LTP (LTPm). Without the secondary Ca(2+) release, only muscarinic LTD (LTDm) was induced. D(-)-2-amino-5-phosphopentanoic acid and intracellular heparin blocked LTPm as well as dendritic Ca(2+) release. A single burst consisting of 3 EPSPs with weak stimulus intensities instead of the SES also induced secondary Ca(2+) release and LTPm. LTPm and LTDm were protein synthesis-dependent. Furthermore, LTPm was confined to specific dendritic compartments and not inducible in distal apical dendrites. Thus, cholinergic activation facilitated selectively compartment-specific induction of late-phase LTP through IP(3)-dependent Ca(2+) release.

Layer- and cell-type-specific tonic GABAergic inhibition of pyramidal neurons in the rat visual cortex.

Tonic inhibition mediated by persistent activation of γ-aminobutyric acidA (GABAA) receptors by ambient GABA plays a crucial role in the regulation of network excitability and neuronal signal processing. Varying degrees in the strength of tonic inhibition were detected across different cell types throughout the brain. Since sensory information flows through cortical layers in a specific order, the characteristics of tonic inhibition in different cortical layers are of interest. Therefore, we examined the properties of tonic inhibition in pyramidal neurons (PyNs) throughout the rat visual cortex. Layer 2/3 PyNs and burst-spiking PyNs in layers 5 and 6 showed prominent tonic GABAA currents. Tonic GABAA currents in layer 4 star PyNs and regular-spiking PyNs in layers 5 and 6 were much weaker. The magnitude of tonic currents correlated well with the inhibition of spike generation. The amplitude of tonic GABAA currents measured with bicuculline and gabazine, the two different GABAA receptor blockers, did not differ. The differences in the expression levels of extrasynaptic GABAA receptors might be the major contributor to the differences in tonic GABAA currents among cell types. Furthermore, α5 subunits might contribute significantly to tonic currents in infragranular burst-spiking PyNs, especially in layer 5. These results suggest that ambient GABA might exert differential effects on the neuronal integration in a layer- and cell-type-specific manner and thus contribute to the processing of sensory properties by selectively tuning the signals flowing through the visual cortex.

Deep Learning for the Pathologic Diagnosis of Hepatocellular Carcinoma, Cholangiocarcinoma, and Metastatic Colorectal Cancer.

Diagnosing primary liver cancers, particularly hepatocellular carcinoma (HCC) and cholangiocarcinoma (CC), is a challenging and labor-intensive process, even for experts, and secondary liver cancers further complicate the diagnosis. Artificial intelligence (AI) offers promising solutions to these diagnostic challenges by facilitating the histopathological classification of tumors using digital whole slide images (WSIs). This study aimed to develop a deep learning model for distinguishing HCC, CC, and metastatic colorectal cancer (mCRC) using histopathological images and to discuss its clinical implications. The WSIs from HCC, CC, and mCRC were used to train the classifiers. For normal/tumor classification, the areas under the curve (AUCs) were 0.989, 0.988, and 0.991 for HCC, CC, and mCRC, respectively. Using proper tumor tissues, the HCC/other cancer type classifier was trained to effectively distinguish HCC from CC and mCRC, with a concatenated AUC of 0.998. Subsequently, the CC/mCRC classifier differentiated CC from mCRC with a concatenated AUC of 0.995. However, testing on an external dataset revealed that the HCC/other cancer type classifier underperformed with an AUC of 0.745. After combining the original training datasets with external datasets and retraining, the classification drastically improved, all achieving AUCs of 1.000. Although these results are promising and offer crucial insights into liver cancer, further research is required for model refinement and validation.

Molecular mechanisms of early growth response protein-1 (EGR-1) expression by quercetin in INS-1 beta-cells.

Early growth response-1 (EGR-1), one of immediate early response genes, is involved in diverse cellular response. We recently reported that quercetin increased catalytic subunit of γ-glutamylcysteine ligase (GCLC) via the interaction of EGR-1 to GCLC promoter in INS-1 beta-cells. Therefore, this study investigated molecular mechanisms of quercetin-induced EGR-1 expression in INS-1 cells. Quercetin significantly induced EGR-1 protein and its mRNA expressions. This induction of EGR-1 was completely blocked by pretreatment with a PKA inhibitor, H89 and partially blocked by a p38 inhibitor, SB203580. Additionally, the siRNA-mediated inhibition of PKAα and p38 resulted in significant reduction of quercetin-induced EGR-1 promoter activity. Also, quercetin-induced EGR-1 protein expression was significantly decreased in the cells transfected with PKAα siRNA. Study using truncated EGR-1 promoter constructs showed that serum response element (SRE) sites, not cAMP response element site, were essential for EGR-1 transcription. However, electrophoretic mobility shift assay showed that quercetin did not affect the band intensity of DNA-protein complex on SRE site of EGR-1 promoter. Also, immune-shift assay using serum response factor (SRF) and phospho-SRF antibodies showed no difference between control and quercetin-treated groups. Collectively, quercetin-induced EGR-1 expression is largely dependent on PKA and partly on p38 MAPK pathway, and SRE sites of EGR-1 promoter are involved in quercetin-induced EGR-1 transcriptional activity.

Layer-specific serotonergic facilitation of IPSC in layer 2/3 pyramidal neurons of the visual cortex.

Serotonin (5-hydroxytryptamine, 5-HT) inhibits the induction of long-term synaptic plasticity in layer 2/3 of the visual cortex at the end of its critical period in rats. However, the cellular and molecular mechanisms remain unclear. Since inhibitory influence is crucial in the induction of synaptic plasticity, the effect of 5-HT on inhibitory transmission was investigated in layer 2/3 pyramidal neurons of the primary visual cortex. The amplitude of inhibitory postsynaptic current (IPSC), but not excitatory postsynaptic current, evoked by stimulation of the underlying layer 4, was increased by ∼20% with a bath application of 5-HT. The amplitude of miniature IPSC was also increased by the application of 5-HT, while the paired-pulse ratio was not changed. The facilitating effect of 5-HT on IPSC was mediated by the activation of 5-HT(2) receptors. An increase in intracellular Ca(2+) via release from inositol 1,4,5-trisphosphate (IP(3))-sensitive stores, which was confirmed by confocal Ca(2+) imaging, and activation of Ca(2+)/calmodulin-dependent kinase II (CaMKII) were involved in the facilitation of IPSC by 5-HT. However, 5-HT failed to facilitate IPSC evoked by the stimulation of layer 1. These results suggest that activation of 5-HT(2) receptors releases intracellular Ca(2+) via IP(3)-sensitive stores, which facilitates GABA(A)ergic transmission via the activation of CaMKII in layer 2/3 pyramidal neurons of the visual cortex in a layer-specific manner. Thus facilitation of inhibitory transmission by 5-HT might be involved in regulating the information flow and the induction of long-term synaptic plasticity, in a pathway-specific manner.

Deep learning-based prediction of molecular cancer biomarkers from tissue slides: A new tool for precision oncology.

Molecular tests are necessary to stratify cancer patients for targeted therapy. However, high cost and technical barriers limit the application of these tests, hindering optimal treatment. Recently, deep learning (DL) has been applied to predict molecular test results from digitized images of tissue slides. Furthermore, treatment response and prognosis can be predicted from tissue slides using DL. In this review, we summarized DL-based studies regarding the prediction of genetic mutation, microsatellite instability, tumor mutational burden, molecular subtypes, gene expression, treatment response, and prognosis directly from hematoxylin- and eosin-stained tissue slides. Although performance needs to be improved, these studies clearly demonstrated the feasibility of DL-based prediction of key molecular features in cancer tissues. With the accumulation of data and technical advances, the performance of the DL system could be improved in the near future. Therefore, we expect that DL could provide cost- and time-effective alternative tools for patient stratification in the era of precision oncology.

Deep Learning-Based Classification of Uterine Cervical and Endometrial Cancer Subtypes from Whole-Slide Histopathology Images.

Uterine cervical and endometrial cancers have different subtypes with different clinical outcomes. Therefore, cancer subtyping is essential for proper treatment decisions. Furthermore, an endometrial and endocervical origin for an adenocarcinoma should also be distinguished. Although the discrimination can be helped with various immunohistochemical markers, there is no definitive marker. Therefore, we tested the feasibility of deep learning (DL)-based classification for the subtypes of cervical and endometrial cancers and the site of origin of adenocarcinomas from whole slide images (WSIs) of tissue slides. WSIs were split into 360 × 360-pixel image patches at 20× magnification for classification. Then, the average of patch classification results was used for the final classification. The area under the receiver operating characteristic curves (AUROCs) for the cervical and endometrial cancer classifiers were 0.977 and 0.944, respectively. The classifier for the origin of an adenocarcinoma yielded an AUROC of 0.939. These results clearly demonstrated the feasibility of DL-based classifiers for the discrimination of cancers from the cervix and uterus. We expect that the performance of the classifiers will be much enhanced with an accumulation of WSI data. Then, the information from the classifiers can be integrated with other data for more precise discrimination of cervical and endometrial cancers.

Recent Applications of Artificial Intelligence from Histopathologic Image-Based Prediction of Microsatellite Instability in Solid Cancers: A Systematic Review.

Cancers with high microsatellite instability (MSI-H) have a better prognosis and respond well to immunotherapy. However, MSI is not tested in all cancers because of the additional costs and time of diagnosis. Therefore, artificial intelligence (AI)-based models have been recently developed to evaluate MSI from whole slide images (WSIs). Here, we aimed to assess the current state of AI application to predict MSI based on WSIs analysis in MSI-related cancers and suggest a better study design for future studies. Studies were searched in online databases and screened by reference type, and only the full texts of eligible studies were reviewed. The included 14 studies were published between 2018 and 2021, and most of the publications were from developed countries. The commonly used dataset is The Cancer Genome Atlas dataset. Colorectal cancer (CRC) was the most common type of cancer studied, followed by endometrial, gastric, and ovarian cancers. The AI models have shown the potential to predict MSI with the highest AUC of 0.93 in the case of CRC. The relatively limited scale of datasets and lack of external validation were the limitations of most studies. Future studies with larger datasets are required to implicate AI models in routine diagnostic practice for MSI prediction.

Deep Learning for Automatic Subclassification of Gastric Carcinoma Using Whole-Slide Histopathology Images.

Histomorphologic types of gastric cancer (GC) have significant prognostic values that should be considered during treatment planning. Because the thorough quantitative review of a tissue slide is a laborious task for pathologists, deep learning (DL) can be a useful tool to support pathologic workflow. In the present study, a fully automated approach was applied to distinguish differentiated/undifferentiated and non-mucinous/mucinous tumor types in GC tissue whole-slide images from The Cancer Genome Atlas (TCGA) stomach adenocarcinoma dataset (TCGA-STAD). By classifying small patches of tissue images into differentiated/undifferentiated and non-mucinous/mucinous tumor tissues, the relative proportion of GC tissue subtypes can be easily quantified. Furthermore, the distribution of different tissue subtypes can be clearly visualized. The patch-level areas under the curves for the receiver operating characteristic curves for the differentiated/undifferentiated and non-mucinous/mucinous classifiers were 0.932 and 0.979, respectively. We also validated the classifiers on our own GC datasets and confirmed that the generalizability of the classifiers is excellent. The results indicate that the DL-based tissue classifier could be a useful tool for the quantitative analysis of cancer tissue slides. By combining DL-based classifiers for various molecular and morphologic variations in tissue slides, the heterogeneity of tumor tissues can be unveiled more efficiently.