Predicting lymph node recurrence in cT1-2N0 tongue squamous cell carcinoma: collaboration between artificial intelligence and pathologists.

Researchers have attempted to identify the factors involved in lymph node recurrence in cT1-2N0 tongue squamous cell carcinoma (SCC). However, studies combining histopathological and clinicopathological information in prediction models are limited. We aimed to develop a highly accurate lymph node recurrence prediction model for clinical stage T1-2, N0 (cT1-2N0) tongue SCC by integrating histopathological artificial intelligence (AI) with clinicopathological information. A dataset from 148 patients with cT1-2N0 tongue SCC was divided into training and test sets. The prediction models were constructed using AI-extracted information from whole slide images (WSIs), human-assessed clinicopathological information, and both combined. Weakly supervised learning and machine learning algorithms were used for WSIs and clinicopathological information, respectively. The combination model utilised both algorithms. Highly predictive patches from the model were analysed for histopathological features. In the test set, the areas under the receiver operating characteristic (ROC) curve for the model using WSI, clinicopathological information, and both combined were 0.826, 0.835, and 0.991, respectively. The highest area under the ROC curve was achieved with the model combining WSI and clinicopathological factors. Histopathological feature analysis showed that highly predicted patches extracted from recurrence cases exhibited significantly more tumour cells, inflammatory cells, and muscle content compared with non-recurrence cases. Moreover, patches with mixed inflammatory cells, tumour cells, and muscle were significantly more prevalent in recurrence versus non-recurrence cases. The model integrating AI-extracted histopathological and human-assessed clinicopathological information demonstrated high accuracy in predicting lymph node recurrence in patients with cT1-2N0 tongue SCC.

Benchmarking Deep Learning-Based Image Retrieval of Oral Tumor Histology.

INTRODUCTION:  Oral tumors necessitate a dependable computer-assisted pathological diagnosis system considering their rarity and diversity. A content-based image retrieval (CBIR) system using deep neural networks has been successfully devised for digital pathology. No CBIR system for oral pathology has been investigated because of the lack of an extensive image database and feature extractors tailored to oral pathology. MATERIALS AND METHODS: This study uses a large CBIR database constructed from 30 categories of oral tumors to compare deep learning methods as feature extractors. RESULTS: The highest average area under the receiver operating characteristic curve (AUC) was achieved by models trained on database images using self-supervised learning (SSL) methods (0.900 with SimCLR and 0.897 with TiCo). The generalizability of the models was validated using query images from the same cases taken with smartphones. When smartphone images were tested as queries, both models yielded the highest mean AUC (0.871 with SimCLR and 0.857 with TiCo). We ensured the retrieved image result would be easily observed by evaluating the top 10 mean accuracies and checking for an exact diagnostic category and its differential diagnostic categories. CONCLUSION: Training deep learning models with SSL methods using image data specific to the target site is beneficial for CBIR tasks in oral tumor histology to obtain histologically meaningful results and high performance. This result provides insight into the effective development of a CBIR system to help improve the accuracy and speed of histopathology diagnosis and advance oral tumor research in the future.

Expression patterns of HNF4α, TTF-1, and SMARCA4 in lung adenocarcinomas: impacts on clinicopathological and genetic features.

INTRODUCTION: HNF4α expression and SMARCA4 loss were thought to be features of non-terminal respiratory unit (TRU)-type lung adenocarcinomas, but their relationships remained unclear. MATERIALS AND METHODS: HNF4α-positive cases among 241 lung adenocarcinomas were stratified based on TTF-1 and SMARCA4 expressions, histological subtypes, and driver mutations. Immunohistochemical analysis was performed using xenograft tumors of lung adenocarcinoma cell lines with high HNF4A expression. RESULT: HNF4α-positive adenocarcinomas(n = 33) were divided into two groups: the variant group(15 mucinous, 2 enteric, and 1 colloid), where SMARCA4 was retained in all cases, and the conventional non-mucinous group(6 papillary, 5 solid, and 4 acinar), where SMARCA4 was lost in 3/15 cases(20%). All variant cases were negative for TTF-1 and showed wild-type EGFR and frequent KRAS mutations(10/18, 56%). The non-mucinous group was further divided into two groups: TRU-type(n = 7), which was positive for TTF-1 and showed predominantly papillary histology(6/7, 86%) and EGFR mutations(3/7, 43%), and non-TRU-type(n = 8), which was negative for TTF-1, showed frequent loss of SMARCA4(2/8, 25%) and predominantly solid histology(4/8, 50%), and never harbored EGFR mutations. Survival analysis of 230 cases based on histological grading and HNF4α expression revealed that HNF4α-positive poorly differentiated (grade 3) adenocarcinoma showed the worst prognosis. Among 39 cell lines, A549 showed the highest level of HNF4A, immunohistochemically HNF4α expression positive and SMARCA4 lost, and exhibited non-mucinous, high-grade morphology in xenograft tumors. CONCLUSION: HNF4α-positive non-mucinous adenocarcinomas included TRU-type and non-TRU-type cases; the latter tended to exhibit the high-grade phenotype with frequent loss of SMARCA4, and A549 was a representative cell line.

High-throughput system for the thermostability analysis of proteins.

Thermal stability of proteins is a primary metric for evaluating their physical properties. Although researchers attempted to predict it using machine learning frameworks, their performance has been dependent on the quality and quantity of published data. This is due to the technical limitation that thermodynamic characterization of protein denaturation by fluorescence or calorimetry in a high-throughput manner has been challenging. Obtaining a melting curve that derives solely from the target protein requires laborious purification, making it far from practical to prepare a hundred or more samples in a single workflow. Here, we aimed to overcome this throughput limitation by leveraging the high protein secretion efficacy of Brevibacillus and consecutive treatment with plate-scale purification methodologies. By handling the entire process of expression, purification, and analysis on a per-plate basis, we enabled the direct observation of protein denaturation in 384 samples within 4 days. To demonstrate a practical application of the system, we conducted a comprehensive analysis of 186 single mutants of a single-chain variable fragment of nivolumab, harvesting the melting temperature (Tm) ranging from -9.3 up to +10.8°C compared to the wild-type sequence. Our findings will allow for data-driven stabilization in protein design and streamlining the rational approaches.

Registered multi-device/staining histology image dataset for domain-agnostic machine learning models.

Variations in color and texture of histopathology images are caused by differences in staining conditions and imaging devices between hospitals. These biases decrease the robustness of machine learning models exposed to out-of-domain data. To address this issue, we introduce a comprehensive histopathology image dataset named PathoLogy Images of Scanners and Mobile phones (PLISM). The dataset consisted of 46 human tissue types stained using 13 hematoxylin and eosin conditions and captured using 13 imaging devices. Precisely aligned image patches from different domains allowed for an accurate evaluation of color and texture properties in each domain. Variation in PLISM was assessed and found to be significantly diverse across various domains, particularly between whole-slide images and smartphones. Furthermore, we assessed the improvement in domain shift using a convolutional neural network pre-trained on PLISM. PLISM is a valuable resource that facilitates the precise evaluation of domain shifts in digital pathology and makes significant contributions towards the development of robust machine learning models that can effectively address challenges of domain shift in histological image analysis.

Association of Genetic Variants with Postoperative Donor Artery Development in Moyamoya Disease: RNF213 and Other Moyamoya Angiopathy-Related Gene Analysis.

Robust postoperative bypass development is a characteristic of moyamoya disease (MMD); however, genetic factors mediating this phenomenon remain incompletely understood. Therefore, we aimed to elucidate the relationship between postoperative donor artery development and genetic variants. We retrospectively enrolled 63 patients (79 hemispheres) who underwent combined revascularization surgery. Postoperative development of the superficial temporal artery (STA), middle meningeal artery, and deep temporal artery (DTA) was assessed using the caliber-change ratio determined from magnetic resonance angiography measurements. We analyzed RNF213 and 36 other moyamoya angiopathy-related genes by whole-exome sequencing and extracted rare or damaging variants. Thirty-five participants carried RNF213 p.Arg4810Lys (all heterozygotes), whereas 5 had RNF213 rare variants (RVs). p.Arg4810Lys was significantly associated with postoperative DTA development, while age at surgery, hypertension, and hyperlipidemia were inversely associated. Multiple regression analysis revealed that age and p.Arg4810Lys held statistical significance (P = 0.044, coefficient - 0.015, 95% confidence interval (CI) - 0.029 to 0.000 and P = 0.001, coefficient 0.670, 95% CI 0.269 to 1.072, respectively). Those with RNF213 RV without p.Arg4810Lys exhibited a significant trend toward poor DTA development (P = 0.001). Hypertension demonstrated a significant positive association with STA development, which remained significant even after multiple regression analysis (P = 0.001, coefficient 0.303, 95% CI 0.123 to 0.482). Following Bonferroni correction for multiple comparisons, targeted analyses of RNF213 and 36 moyamoya angiopathy-related genes showed a significant association of only RNF213 p.Arg4810Lys with favorable DTA development (P = 0.001). A comprehensive analysis of RNF213, considering both p.Arg4810Lys and RVs, may provide a clearer prediction of postoperative DTA development.

Engineering CD3/CD137 Dual Specificity into a DLL3-Targeted T-Cell Engager Enhances T-Cell Infiltration and Efficacy against Small-Cell Lung Cancer.

Small-cell lung cancer (SCLC) is an aggressive cancer for which immune checkpoint inhibitors (ICI) have had only limited success. Bispecific T-cell engagers are promising therapeutic alternatives for ICI-resistant tumors, but not all patients with SCLC are responsive. Herein, to integrate CD137 costimulatory function into a T-cell engager format and thereby augment therapeutic efficacy, we generated a CD3/CD137 dual-specific Fab and engineered a DLL3-targeted trispecific antibody (DLL3 trispecific). The CD3/CD137 dual-specific Fab was generated to competitively bind to CD3 and CD137 to prevent DLL3-independent cross-linking of CD3 and CD137, which could lead to systemic T-cell activation. We demonstrated that DLL3 trispecific induced better tumor growth control and a marked increase in the number of intratumoral T cells compared with a conventional DLL3-targeted bispecific T-cell engager. These findings suggest that DLL3 trispecific can exert potent efficacy by inducing concurrent CD137 costimulation and provide a promising therapeutic option for SCLC.

Spatial intratumor heterogeneity of programmed death-ligand 1 expression predicts poor prognosis in resected non-small cell lung cancer.

BACKGROUND: We quantified the pathological spatial intratumor heterogeneity of programmed death-ligand 1 (PD-L1) expression and investigated its relevance to patient outcomes in surgically resected non-small cell lung carcinoma (NSCLC). METHODS: This study enrolled 239 consecutive surgically resected NSCLC specimens of pathological stage IIA-IIIB. To characterize the spatial intratumor heterogeneity of PD-L1 expression in NSCLC tissues, we developed a mathematical model based on texture image analysis and determined the spatial heterogeneity index of PD-L1 for each tumor. The correlation between the spatial heterogeneity index of PD-L1 values and clinicopathological characteristics, including prognosis, was analyzed. Furthermore, an independent cohort of 70 cases was analyzed for model validation. RESULTS: Clinicopathological analysis showed correlations between high spatial heterogeneity index of PD-L1 values and histological subtype (squamous cell carcinoma; P < .001) and vascular invasion (P = .004). Survival analysis revealed that patients with high spatial heterogeneity index of PD-L1 values presented a significantly worse recurrence-free rate than those with low spatial heterogeneity index of PD-L1 values (5-year recurrence-free survival [RFS] = 26.3% vs 47.1%, P < .005). The impact of spatial heterogeneity index of PD-L1 on cancer survival rates was verified through validation in an independent cohort. Additionally, high spatial heterogeneity index of PD-L1 values were associated with tumor recurrence in squamous cell carcinoma (5-year RFS = 29.2% vs 52.8%, P < .05) and adenocarcinoma (5-year RFS = 19.6% vs 43.0%, P < .01). Moreover, we demonstrated that a high spatial heterogeneity index of PD-L1 value was an independent risk factor for tumor recurrence. CONCLUSIONS: We presented an image analysis model to quantify the spatial intratumor heterogeneity of protein expression in tumor tissues. This model demonstrated that the spatial intratumor heterogeneity of PD-L1 expression in surgically resected NSCLC predicts poor patient outcomes.

Extracting interpretable features for pathologists using weakly supervised learning to predict p16 expression in oropharyngeal cancer.

One drawback of existing artificial intelligence (AI)-based histopathological prediction models is the lack of interpretability. The objective of this study is to extract p16-positive oropharyngeal squamous cell carcinoma (OPSCC) features in a form that can be interpreted by pathologists using AI model. We constructed a model for predicting p16 expression using a dataset of whole-slide images from 114 OPSCC biopsy cases. We used the clustering-constrained attention-based multiple-instance learning (CLAM) model, a weakly supervised learning approach. To improve performance, we incorporated tumor annotation into the model (Annot-CLAM) and achieved the mean area under the receiver operating characteristic curve of 0.905. Utilizing the image patches on which the model focused, we examined the features of model interest via histopathologic morphological analysis and cycle-consistent adversarial network (CycleGAN) image translation. The histopathologic morphological analysis evaluated the histopathological characteristics of image patches, revealing significant differences in the numbers of nuclei, the perimeters of the nuclei, and the intercellular bridges between p16-negative and p16-positive image patches. By using the CycleGAN-converted images, we confirmed that the sizes and densities of nuclei are significantly converted. This novel approach improves interpretability in histopathological morphology-based AI models and contributes to the advancement of clinically valuable histopathological morphological features.