Assessment of associations between clinical and immune microenvironmental factors and tumor mutation burden in resected nonsmall cell lung cancer by applying machine learning to whole-slide images

A machine learning-based radiomics model for prediction of tumor mutation burden in gastric cancer

Purpose: To evaluate the potential of machine learning (ML)-based radiomics approach for predicting tumor mutation burden (TMB) in gastric cancer (GC). Methods: The contrast enhanced CT (CECT) images with corresponding clinical information of 256 GC patients were retrospectively collected. Patients were separated into training set (n = 180) and validation set (n = 76). A total of 3,390 radiomics features were extracted from three phases images of CECT. The least absolute shrinkage and selection operator (LASSO) model was used for feature screening. Seven machine learning (ML) algorithms were employed to find the optimal classifier. The predictive ability of radiomics model (RM) was evaluated with receiver operating characteristic. The correlation between RM and TMB values was evaluated using Spearman's correlation coefficient. The explainability of RM was assessed by the Shapley Additive explanations (SHAP) method. Results: Logistic regression algorithm was chosen for model construction. The RM showed good predictive ability of TMB status with AUCs of 0.89 [95% confidence interval (CI): 0.85-0.94] and 0.86 (95% CI: 0.74-0.98) in the training and validation sets. The correlation analysis revealed a good correlation between RM and TMB levels (correlation coefficient: 0.62, p < 0.001). The RM also showed favorable and stable predictive accuracy within the cutoff value range 6-16 mut/Mb in both sets. Conclusion: The ML-based RM offered a promising image biomarker for predicting TMB status in GC patients.

Predictive value of 18 F-FDG PET/CT and serum tumor markers for tumor mutational burden in patients with non-small cell lung cancer

PURPOSE

To investigate the correlations between metabolic parameters (MPs) of 18 F-fluorodeoxyglucose (FDG) uptake on positron emission tomography/computed tomography (PET/CT), serum tumor markers (STMs), and tumor mutational burden (TMB) in patients with non-small cell lung cancer (NSCLC).

MATERIALS AND METHODS

In this retrospective study, we enrolled 129 patients with NSCLC (males, 78; females, 51) who underwent baseline TMB and STM tests and 18 F-FDG PET/CT scans before treatment between March 2018 and September 2022. Patients were categorized into TMB-high (TMB ≥10 mutations/Mb; n = 27 [20.9%]) and non-TMB-high (TMB <10 mutations/Mb; n = 102 [79.1%]) groups. Binary logistic regression analyses were performed to determine independent predictors of TMB-high. Univariate and multivariate linear regression analyses were performed to determine independent predictors of TMB level on a log scale. Subgroup analyses for adenocarcinoma (ADC), ADC with EGFR+, ADC with EGFR-, and squamous cell carcinoma (SCC) were performed.

RESULTS

For ADC, all MPs (SULpeak , SULmax , SULmean , MTV, and TLG) were significantly higher in the TMB-high group than the non-TMB-high group; smoker (odds ratio [OR] = 27.08, p = 0.018), EGFR+ (OR = 0.03, p = 0.033), KRAS+ (OR = 7.98, p = 0.083), high CEA (OR = 33.56, p = 0.029), and high CA125 (OR = 13.68, p = 0.030) were independent predictors of TMB-high; and all MPs showed significant positive linear correlations with TMB on a log scale, with SULpeak as an independent predictor. However, no significant correlation was observed for SCC.

CONCLUSION

MPs and STMs can predict the TMB level for patients with ADC, and may serve as potential substitutes for TMB with increased value and easy implementation in guiding immunotherapy through noninvasive methods.

Prognostic significance of tumor microenvironment assessed by machine learning algorithm in surgically resected non-small cell lung cancer

BACKGROUND

A methodology to assess the immune microenvironment (IME) of non-small cell lung cancer (NSCLC) has not been established, and the prognostic impact of IME factors is not yet clear.

AIMS

This study aimed to assess the IME factors and evaluate their prognostic values.

METHODS AND RESULTS

We assessed CD8+ tumor-infiltrating lymphocyte (TIL) density, forkhead box protein P3+ (Foxp3+ ) TIL density, and programmed death receptor ligand-1 (PD-L1) tumor proportion score (TPS) using a machine-learning algorithm in whole-slide imaging (WSI). We dichotomized patients according to TIL density or TPS and compared their clinical outcomes. Between September 2014 and September 2015, 165 patients with NSCLC were enrolled in the study. We assessed IME factors in the epithelium, stroma, and their combination. An improvement in disease-free survival (DFS) was observed in the high CD8+ TIL density group in the epithelium, stroma, and the combination of both. Moreover, the group with high PD-L1 TPS in the epithelium showed better DFS than that with low PD-L1 TPS. In the multivariate analysis, the CD8+ TIL density in the combination of epithelium and stroma and PD-L1 TPS in the epithelium were independent prognostic factors (hazard ratio [HR] = 0.43; 95% confidence interval [CI] = 0.26-0.72; p = .001, HR = 0.49; 95% CI = 0.30-0.81; p = .005, respectively).

CONCLUSION

Our approach demonstrated that the IME factors are related to survival in patients with NSCLC. The quantitative assessment of IME factors enables to discriminate patients with high risk of recurrence, who can be the candidates for adjuvant therapy. Assessing the CD8+ TIL density in the combination of epithelium and stroma might be more useful than their individual assessment because it is a simple and time-saving analysis of TILs in WSI.

The artificial intelligence and machine learning in lung cancer immunotherapy

Since the past decades, more lung cancer patients have been experiencing lasting benefits from immunotherapy. It is imperative to accurately and intelligently select appropriate patients for immunotherapy or predict the immunotherapy efficacy. In recent years, machine learning (ML)-based artificial intelligence (AI) was developed in the area of medical-industrial convergence. AI can help model and predict medical information. A growing number of studies have combined radiology, pathology, genomics, proteomics data in order to predict the expression levels of programmed death-ligand 1 (PD-L1), tumor mutation burden (TMB) and tumor microenvironment (TME) in cancer patients or predict the likelihood of immunotherapy benefits and side effects. Finally, with the advancement of AI and ML, it is believed that "digital biopsy" can replace the traditional single assessment method to benefit more cancer patients and help clinical decision-making in the future. In this review, the applications of AI in PD-L1/TMB prediction, TME prediction and lung cancer immunotherapy are discussed.

The efficacy of a machine learning algorithm for assessing tumour components as a prognostic marker of surgically resected stage IA lung adenocarcinoma

BACKGROUND

The importance of the stromal components in tumour progression has been discussed widely, but their prognostic role in small size tumours with lepidic components is not fully understood. Applying digital tissue image analysis to whole-slide imaging may enhance the accuracy and reproducibility of pathological assessment. This study aimed to evaluate the prognostic value of tumour components of lung adenocarcinoma by measuring the dimensions of the tumour consisting elements separately, using a machine learning algorithm.

METHODS

Between September 2002 and December 2016, 317 patients with surgically resected, pathological stage IA adenocarcinoma with lepidic components were analysed. We assessed the whole tumour area, including the lepidic components, and measured the epithelium, collagen, elastin areas and alveolar air space. We analysed the prognostic impact of each tumour component.

RESULTS

The dimensions of the epithelium and collagen areas were independent significant risk factors for recurrence-free survival (hazard ratio, 8.38; 95% confidence interval, 1.14-61.88; P = 0.037, and hazard ratio, 2.58; 95% confidence interval, 1.14-5.83; P = 0.022, respectively). According to the subgroup analysis when combining the epithelium and collagen areas as risk factors, patients with tumours consisting of both large epithelium and collagen areas showed significantly poor prognoses (P = 0.002).

CONCLUSIONS

We assessed tumour components using a machine learning algorithm to stratify the post-operative prognosis of surgically resected stage IA adenocarcinomas. This method might guide the selection of patients with a high risk of recurrence.

Quantitative Assessment of Tumor-Infiltrating Lymphocytes Using Machine Learning Predicts Survival in Muscle-Invasive Bladder Cancer

(1) Purpose: Although assessment of tumor-infiltrating lymphocytes (TILs) has been acknowledged to have important predictive prognostic value in muscle-invasive bladder cancer (MIBC), it is limited by inter- and intra-observer variability, hampering widespread clinical application. We aimed to evaluate the prognostic value of quantitative TILs score based on a machine learning (ML) algorithm to identify MIBC patients who might benefit from immunotherapy or the de-escalation of therapy. (2) Methods: We constructed an artificial neural network classifier for tumor cells, lymphocytes, stromal cells, and “ignore” cells from hematoxylin-and-eosin-stained slide images using the QuPath open source software. We defined four unique TILs variables based on ML to analyze TILs measurements. Pathological slide images from 133 MIBC patients were retrospectively collected as the discovery set to determine the optimal association of ML-read TILs variables with patient survival outcomes. For validation, we evaluated an independent external validation set consisting of 247 MIBC patients. (3) Results: We found that all four TILs variables had significant prognostic associations with survival outcomes in MIBC patients (p < 0.001 for all comparisons), with higher TILs score being associated with better prognosis. Univariate and multivariate Cox regression analyses demonstrated that electronic TILs (eTILs) variables were independently associated with overall survival after adjustment for clinicopathological factors including age, sex, and pathological stage (p < 0.001 for all analyses). Results analyzed in different subgroups showed that the eTILs variable was a strong prognostic factor that was not redundant with pre-existing clinicopathological features (p < 0.05 for all analyses). (4) Conclusion: ML-driven cell classifier-defined TILs variables were robust and independent prognostic factors in two independent cohorts of MIBC patients. eTILs have the potential to identify a subset of high-risk stage II or stage III-IV MIBC patients who might benefit from adjuvant immunotherapy.

Artificial Intelligence-Powered Prediction of ALK Gene Rearrangement in Patients With Non-Small-Cell Lung Cancer

PURPOSE

Several studies reported the possibility of predicting genetic abnormalities in non-small-cell lung cancer by deep learning (DL). However, there are no data of predicting ALK gene rearrangement (ALKr) using DL. We evaluated the ALKr predictability using the DL platform.

MATERIALS AND METHODS

We selected 66 ALKr-positive cases and 142 ALKr-negative cases, which were diagnosed by ALKr immunohistochemical staining in our institution from January 2009 to March 2019. We generated virtual slide of 300 slides (150 ALKr-positive slides and 150 ALKr-negative slides) using NanoZoomer. HALO-AI was used to analyze the whole-slide imaging data, and the DenseNet network was used to build the learning model. Of the 300 slides, we randomly assigned 172 slides to the training cohort and 128 slides to the test cohort to ensure no duplication of cases. In four resolutions (16.0/4.0/1.0/0.25 μm/pix), ALKr prediction models were built in the training cohort and ALKr prediction performance was evaluated in the test cohort. We evaluated the diagnostic probability of ALKr by receiver operating characteristic analysis in each ALKr probability threshold (50%, 60%, 70%, 80%, 90%, and 95%). We expected the area under the curve to be 0.64-0.85 in the model of a previous study. Furthermore, in the test cohort data, an expert pathologist also evaluated the presence of ALKr by hematoxylin and eosin staining on whole-slide imaging.

RESULTS

The maximum area under the curve was 0.73 (50% threshold: 95% CI, 0.65 to 0.82) in the resolution of 1.0 μm/pix. In this resolution, with an ALKr probability of 50% threshold, the sensitivity and specificity were 73% and 73%, respectively. The expert pathologist's sensitivity and specificity in the same test cohort were 13% and 94%.

CONCLUSION

The ALKr prediction by DL was feasible. Further study should be addressed to improve accuracy of ALKr prediction.

Assessing dynamic metabolic heterogeneity in non-small cell lung cancer patients via ultra-high sensitivity total-body [18F]FDG PET/CT imaging: quantitative analysis of [18F]FDG uptake in primary tumors and metastatic lymph nodes

PURPOSE

This study aimed to quantitatively assess [¹⁸F]FDG uptake in primary tumor (PT) and metastatic lymph node (mLN) in newly diagnosed non-small cell lung cancer (NSCLC) using the total-body [¹⁸F]FDG PET/CT and to characterize the dynamic metabolic heterogeneity of NSCLC.

METHODS

The 60-min dynamic total-body [¹⁸F]FDG PET/CT was performed before treatment. The PTs and mLNs were manually delineated. An unsupervised K-means classification method was used to cluster patients based on the imaging features of PTs. The metabolic features, including Patlak-Ki, Patlak-Intercept, SUV_mean, metabolic tumor volume (MTV), total lesion glycolysis (TLG), and textural features, were extracted from PTs and mLNs. The targeted next-generation sequencing of tumor-associated genes was performed. The expression of Ki67, CD3, CD8, CD34, CD68, and CD163 in PTs was determined by immunohistochemistry.

RESULTS

A total of 30 patients with stage IIIA-IV NSCLC were enrolled. Patients were divided into fast dynamic FDG metabolic group (F-DFM) and slow dynamic FDG metabolic group (S-DFM) by the unsupervised K-means classification of PTs. The F-DFM group showed significantly higher Patlak-Ki (P < 0.001) and SUV_mean (P < 0.001) of PTs compared with the S-DFM group, while no significant difference was observed in Patlak-Ki and SUV_mean of mLNs between the two groups. The texture analysis indicated that PTs in the S-DFM group were more heterogeneous in FDG uptake than those in the F-DFM group. Higher T cells (CD3⁺/CD8⁺) and macrophages (CD68⁺/CD163⁺) infiltration in the PTs were observed in the F-DFM group. No significant difference was observed in tumor mutational burden between the two groups.

CONCLUSION

The dynamic total-body [¹⁸F]FDG PET/CT stratified NSCLC patients into the F-DFM and S-DFM groups, based on Patlak-Ki and SUV_mean of PTs. PTs in the F-DFM group seemed to be more homogenous in terms of [¹⁸F]FDG uptake than those in the S-DFM group. The higher infiltrations of T cells and macrophages were observed in the F-DFM group, which suggested a potential benefit from immunotherapy.

Dynamic monitoring serum tumor markers to predict molecular features of EGFR-mutated lung cancer during targeted therapy

To reveal the correlation of dynamic serum tumor markers (STMs) and molecular features of epidermal growth factor receptor‐mutated (EGFR‐mutated) lung cancer during targeted therapy, we retrospectively reviewed 303 lung cancer patients who underwent dynamic STM tests [neuron‐specific enolase (NSE), carcinoembryonic antigen (CEA), carbohydrate antigen 125 (CA125), carbohydrate antigen 153 (CA153), the soluble fragment of cytokeratin 19 (CYFRA21‐1), and squamous cell carcinoma antigen (SCC)] and circulating tumor DNA (ctDNA) testing with a panel covering 168 genes. At baseline, patients with EGFR mutation trended to have abnormal CEA, abnormal CA153, and normal SCC levels. Additionally, patients with Thr790Met (T790M) mutation were more likely to have abnormal CEA levels than patients without T790M mutation. Among patients with secondary resistance to EGFR tyrosine kinase inhibitors (TKI), the dynamic STMs showed a descending trend in the responsive stage and a rising trend in the resistant stage. However, the changing slopes differed between T790M subgroup and the non‐T790M subgroup in individual STMs. Our study demonstrated that the combination of baseline levels and variations of STMs (including the responsive stage and resistant stage) can be suggestive of secondary EGFR‐T790M mutation [area under the curve (AUC) = 0.897] and that changing trends of STMs (within 8 weeks after initiating the TKI therapy) can be potential predictors for the clearance of EGFR ctDNA [AUC = 0.871]. In conclusion, dynamic monitoring STMs can help to predict the molecular features of EGFR‐mutated lung cancer during targeted therapy.

A whole-slide image (WSI)-based immunohistochemical feature prediction system improves the subtyping of lung cancer

BACKGROUND

Clinically, accurate pathological diagnosis is often challenged by insufficient tissue amounts and the unaffordability of additional immunohistochemical or genetic tests; thus, there is an urgent need for a universal approach to improve the subtyping of lung cancer without the above limitations. Here we aimed to develop a deep learning system to predict the immunohistochemistry (IHC) phenotype directly from whole-slide images (WSIs) to improve the subtyping of lung cancer from surgical resection and biopsy specimens.

METHODS

A total of 1914 patients with lung cancer from three independent hospitals in China were enrolled for WSI-based immunohistochemical feature prediction system (WIFPS) development and validation.

RESULTS

The WIFPS could directly predict the IHC status of nine subtype-specific biomarkers, including CK7, TTF-1, Napsin A, CK5/6, P63, P40, CD56, Synaptophysin, and Chromogranin A, achieving average areas under the curve (AUCs) of 0.912, 0.906, and 0.888 and overall diagnostic accuracies of 0.925, 0.941, and 0.887 in the validation datasets of total, external surgical resection specimens and biopsy specimens, respectively. The histological subtyping performance of the WIFPS remained comparable with that of general pathologists (GPs), with Cohen's kappa values ranging from 0.7646 to 0.8282. Furthermore, the WIFPS could be trained to not only predict the IHC status of anaplastic lymphoma kinase (ALK), programmed death-1 (PD-1), and programmed death ligand 1 (PD-L1), but also predict EGFR and KRAS mutation status, with AUCs from 0.525 to 0.917, as detected in separate populations.

CONCLUSIONS

In this study, the WIFPS showed its proficiency as a useful complement to traditional histologic subtyping for integrated immunohistochemical spectrum prediction as well as potential in the detection of gene mutations.

A Study of the Recent Trends of Immunology: Key Challenges, Domains, Applications, Datasets, and Future Directions

The human immune system is very complex. Understanding it traditionally required specialized knowledge and expertise along with years of study. However, in recent times, the introduction of technologies such as AIoMT (Artificial Intelligence of Medical Things), genetic intelligence algorithms, smart immunological methodologies, etc., has made this process easier. These technologies can observe relations and patterns that humans do and recognize patterns that are unobservable by humans. Furthermore, these technologies have also enabled us to understand better the different types of cells in the immune system, their structures, their importance, and their impact on our immunity, particularly in the case of debilitating diseases such as cancer. The undertaken study explores the AI methodologies currently in the field of immunology. The initial part of this study explains the integration of AI in healthcare and how it has changed the face of the medical industry. It also details the current applications of AI in the different healthcare domains and the key challenges faced when trying to integrate AI with healthcare, along with the recent developments and contributions in this field by other researchers. The core part of this study is focused on exploring the most common classifications of health diseases, immunology, and its key subdomains. The later part of the study presents a statistical analysis of the contributions in AI in the different domains of immunology and an in-depth review of the machine learning and deep learning methodologies and algorithms that can and have been applied in the field of immunology. We have also analyzed a list of machine learning and deep learning datasets about the different subdomains of immunology. Finally, in the end, the presented study discusses the future research directions in the field of AI in immunology and provides some possible solutions for the same.

Integrating Histologic and Genomic Characteristics to Predict Tumor Mutation Burden of Early-Stage Non-Small-Cell Lung Cancer

Tumor mutation burden (TMB) serves as an effective biomarker predicting efficacy of mono-immunotherapy for non-small cell lung cancer (NSCLC). Establishing a precise TMB predicting model is essential to select which populations are likely to respond to immunotherapy or prognosis and to maximize the benefits of treatment. In this study, available Formalin-fixed paraffin embedded tumor tissues were collected from 499 patients with NSCLC. Targeted sequencing of 636 cancer related genes was performed, and TMB was calculated. Distribution of TMB was significantly (p < 0.001) correlated with sex, clinical features (pathological/histological subtype, pathological stage, lymph node metastasis, and lympho-vascular invasion). It was also significantly (p < 0.001) associated with mutations in genes like TP53, EGFR, PIK3CA, KRAS, EPHA3, TSHZ3, FAT3, NAV3, KEAP1, NFE2L2, PTPRD, LRRK2, STK11, NF1, KMT2D, and GRIN2A. No significant correlations were found between TMB and age, neuro-invasion (p = 0.125), and tumor location (p = 0.696). Patients with KRAS p.G12 mutations and FAT3 missense mutations were associated (p < 0.001) with TMB. TP53 mutations also influence TMB distribution (P < 0.001). TMB was reversely related to EGFR mutations (P < 0.001) but did not differ by mutation types. According to multivariate logistic regression model, genomic parameters could effectively construct model predicting TMB, which may be improved by introducing clinical information. Our study demonstrates that genomic together with clinical features yielded a better reliable model predicting TMB-high status. A simplified model consisting of less than 20 genes and couples of clinical parameters were sought to be useful to provide TMB status with less cost and waiting time.

Role of Tumor Mutation Burden Analysis in Detecting Lynch Syndrome in Precision Medicine: Analysis of 2,501 Japanese Cancer Patients

BACKGROUND

Tumor mutation burden (TMB) is the total exonic mutation count per megabase of tumor DNA. Recent advances in precision medicine occasionally detect Lynch syndrome (LS) by germline sequencing for mismatch-repair (g.MMR) genes but not using TMB. The current study analyzes the utility of TMB in detecting LS.

METHODS

Whole-exome sequencing (ion-semiconductor sequencing) was performed for somatic and germline DNA from 2,501 various cancer patients to detect TMB and g.MMR sequencing. MMR IHC was conducted when high TMB (≥10) was detected in LS-related cancers with an additional condition of wild-type BRAF in colorectal cancers. Target sequencing and multiplex ligation-dependent probe amplification (MLPA) were further performed for g.MMR genes in MMR-deficient cancers (TMB-based g.MMR target sequencing). We compared universal sequencing and TMB-based target sequencing in their sensitivity for detecting LS.

RESULTS

LS was detected in 16 (0.6%) of the 2,501 patients: 1.1% (9/826) of colorectal cancer patients, 16.2% (6/37) of endometrial cancer patients, and 14.3% (1/7) of small intestine cancer patients. TMB-based g.MMR target sequencing (81.3%) showed superior sensitivity for detecting LS than universal g.MMR sequencing (56.3%; P = 0.127) but missed 3 LS patients (1 with a low-TMB cancer, 1 with a BRAF-mutant colorectal cancer, and 1 with an MMR-proficient cancer). Ion-semiconductor sequencing could detect single-nucleotide substitutions but not large deletions. POL-mutated cancers showed extremely high TMBs (48.4-749.2).

CONCLUSIONS

g.MMR target sequencing, combined with TMB, somatic BRAF mutation, and MMR IHC is an effective strategy for detecting LS.

IMPACT

TMB can be a biomarker for detecting LS in precision medicine.

Assessment of associations between clinical and immune microenvironmental factors and tumor mutation burden in resected nonsmall cell lung cancer by applying machine learning to whole-slide images

BACKGROUND

It is unclear whether clinical factors and immune microenvironment (IME) factors are associated with tumor mutation burden (TMB) in patients with nonsmall cell lung cancer (NSCLC).

MATERIALS AND METHODS

We assessed TMB in surgical tumor specimens by performing whole exome sequencing. IME profiles, including PD-L1 tumor proportion score (TPS), stromal CD8 tumor-infiltrating lymphocyte (TIL) density, and stromal Foxp3 TIL density, were quantified by digital pathology using a machine learning algorithm. To detect factors associated with TMB, clinical data, and IME factors were assessed by means of a multiple regression model.

RESULTS

We analyzed tumors from 200 of the 246 surgically resected NSCLC patients between September 2014 and September 2015. Patient background: median age (range) 70 years (39-87); male 37.5%; smoker 27.5%; pathological stage (p-stage) I/II/III, 63.5/22.5/14.0%; histological type Ad/Sq, 77.0/23.0%; primary tumor location upper/lower, 58.5/41.5%; median PET SUV 7.5 (0.86-29.8); median serum CEA (sCEA) level 3.4 ng/mL (0.5-144.3); median serum CYFRA 21-1 (sCYFRA) level 1.2 ng/mL (1.0-38.0); median TMB 2.19/ Mb (0.12-64.38); median PD-L1 TPS 15.1% (0.09-77.4); median stromal CD8 TIL density 582.1/mm2 (120.0-4967.6);, and median stromal Foxp3 TIL density 183.7/mm2 (6.3-544.0). The multiple regression analysis identified three factors associated with higher TMB: smoking status: smoker, increase PET SUV, and sCEA level: >5 ng/mL (P < .001, P < .001, and P = .006, respectively).

CONCLUSIONS

The IME factors assessed were not associated with TMB, but our findings showed that, in addition to smoking, PET SUV and sCEA levels may be independent predictors of TMB. TMB and IME factors are independent factors in resected NSCLC.