Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

General General

A comparison of machine learning models and Cox proportional hazards models regarding their ability to predict the risk of gastrointestinal cancer based on metabolic syndrome and its components.

In Frontiers in oncology

BACKGROUND : Little is known about applying machine learning (ML) techniques to identify the important variables contributing to the occurrence of gastrointestinal (GI) cancer in epidemiological studies. We aimed to compare different ML models to a Cox proportional hazards (CPH) model regarding their ability to predict the risk of GI cancer based on metabolic syndrome (MetS) and its components.

METHODS : A total of 41,837 participants were included in a prospective cohort study. Incident cancer cases were identified by following up with participants until December 2019. We used CPH, random survival forest (RSF), survival trees (ST), gradient boosting (GB), survival support vector machine (SSVM), and extra survival trees (EST) models to explore the impact of MetS on GI cancer prediction. We used the C-index and integrated Brier score (IBS) to compare the models.

RESULTS : In all, 540 incident GI cancer cases were identified. The GB and SSVM models exhibited comparable performance to the CPH model concerning the C-index (0.725). We also recorded a similar IBS for all models (0.017). Fasting glucose and waist circumference were considered important predictors.

CONCLUSIONS : Our study found comparably good performance concerning the C-index for the ML models and CPH model. This finding suggests that ML models may be considered another method for survival analysis when the CPH model's conditions are not satisfied.

Tran Tao Thi, Lee Jeonghee, Gunathilake Madhawa, Kim Junetae, Kim Sun-Young, Cho Hyunsoon, Kim Jeongseon

2023

Korea, gastrointestinal cancer, machine learning, metabolic syndrome, prospective cohort study

Radiology Radiology

A CT based radiomics nomogram for differentiation between focal-type autoimmune pancreatitis and pancreatic ductal adenocarcinoma.

In Frontiers in oncology

OBJECTIVES : The purpose of this study was to develop and validate an CT-based radiomics nomogram for the preoperative differentiation of focal-type autoimmune pancreatitis from pancreatic ductal adenocarcinoma.

METHODS : 96 patients with focal-type autoimmune pancreatitis and pancreatic ductal adenocarcinoma have been enrolled in the study (32 and 64 cases respectively). All cases have been confirmed by imaging, clinical follow-up and/or pathology. The imaging data were considered as: 70% training cohort and 30% test cohort. Pancreatic lesions have been manually delineated by two radiologists and image segmentation was performed to extract radiomic features from the CT images. Independent-sample T tests and LASSO regression were used for feature selection. The training cohort was classified using a variety of machine learning-based classifiers, and 5-fold cross-validation has been performed. The classification performance was evaluated using the test cohort. Multivariate logistic regression analysis was then used to develop a radiomics nomogram model, containing the CT findings and Rad-Score. Calibration curves have been plotted showing the agreement between the predicted and actual probabilities of the radiomics nomogram model. Different patients have been selected to test and evaluate the model prediction process. Finally, receiver operating characteristic curves and decision curves were plotted, and the radiomics nomogram model was compared with a single model to visually assess its diagnostic ability.

RESULTS : A total of 158 radiomics features were extracted from each image. 7 features were selected to construct the radiomics model, then a variety of classifiers were used for classification and multinomial logistic regression (MLR) was selected to be the optimal classifier. Combining CT findings with radiomics model, a prediction model based on CT findings and radiomics was finally obtained. The nomogram model showed a good sensitivity and specificity with AUCs of 0.87 and 0.83 in training and test cohorts, respectively. The areas under the curve and decision curve analysis showed that the radiomics nomogram model may provide better diagnostic performance than the single model and achieve greater clinical net benefits than the CT finding model and radiomics signature model individually.

CONCLUSIONS : The CT image-based radiomics nomogram model can accurately distinguish between focal-type autoimmune pancreatitis and pancreatic ductal adenocarcinoma patients and provide additional clinical benefits.

Lu Jia, Jiang Nannan, Zhang Yuqing, Li Daowei

2023

differential, focal-type autoimmune pancreatitis, machine learning, pancreatic ductal adenocarcinoma, radiomics

Surgery Surgery

Identification of the mitophagy-related diagnostic biomarkers in hepatocellular carcinoma based on machine learning algorithm and construction of prognostic model.

In Frontiers in oncology

BACKGROUND AND AIMS : As a result of increasing numbers of studies most recently, mitophagy plays a vital function in the genesis of cancer. However, research on the predictive potential and clinical importance of mitophagy-related genes (MRGs) in hepatocellular carcinoma (HCC) is currently lacking. This study aimed to uncover and analyze the mitophagy-related diagnostic biomarkers in HCC using machine learning (ML), as well as to investigate its biological role, immune infiltration, and clinical significance.

METHODS : In our research, by using Least absolute shrinkage and selection operator (LASSO) regression and support vector machine- (SVM-) recursive feature elimination (RFE) algorithm, six mitophagy genes (ATG12, CSNK2B, MTERF3, TOMM20, TOMM22, and TOMM40) were identified from twenty-nine mitophagy genes, next, the algorithm of non-negative matrix factorization (NMF) was used to separate the HCC patients into cluster A and B based on the six mitophagy genes. And there was evidence from multi-analysis that cluster A and B were associated with tumor immune microenvironment (TIME), clinicopathological features, and prognosis. After then, based on the DEGs (differentially expressed genes) between cluster A and cluster B, the prognostic model (riskScore) of mitophagy was constructed, including ten mitophagy-related genes (G6PD, KIF20A, SLC1A5, TPX2, ANXA10, TRNP1, ADH4, CYP2C9, CFHR3, and SPP1).

RESULTS : This study uncovered and analyzed the mitophagy-related diagnostic biomarkers in HCC using machine learning (ML), as well as to investigate its biological role, immune infiltration, and clinical significance. Based on the mitophagy-related diagnostic biomarkers, we constructed a prognostic model(riskScore). Furthermore, we discovered that the riskScore was associated with somatic mutation, TIME, chemotherapy efficacy, TACE and immunotherapy effectiveness in HCC patients.

CONCLUSION : Mitophagy may play an important role in the development of HCC, and further research on this issue is necessary. Furthermore, the riskScore performed well as a standalone prognostic marker in terms of accuracy and stability. It can provide some guidance for the diagnosis and treatment of HCC patients.

Tu Dao-Yuan, Cao Jun, Zhou Jie, Su Bing-Bing, Wang Shun-Yi, Jiang Guo-Qing, Jin Sheng-Jie, Zhang Chi, Peng Rui, Bai Dou-Sheng

2023

bioinformatics, hepatocelluar carcinoma, machine learning, mitophagy, prognostic model, time

Radiology Radiology

Self-supervised contrastive learning using CT images for PD-1/PD-L1 expression prediction in hepatocellular carcinoma.

In Frontiers in oncology

BACKGROUND AND PURPOSE : Programmed cell death protein-1 (PD-1) and programmed cell death-ligand-1 (PD-L1) expression status, determined by immunohistochemistry (IHC) of specimens, can discriminate patients with hepatocellular carcinoma (HCC) who can derive the most benefits from immune checkpoint inhibitor (ICI) therapy. A non-invasive method of measuring PD-1/PD-L1 expression is urgently needed for clinical decision support.

MATERIALS AND METHODS : We included a cohort of 87 patients with HCC from the West China Hospital and analyzed 3094 CT images to develop and validate our prediction model. We propose a novel deep learning-based predictor, Contrastive Learning Network (CLNet), which is trained with self-supervised contrastive learning to better extract deep representations of computed tomography (CT) images for the prediction of PD-1 and PD-L1 expression.

RESULTS : Our results show that CLNet exhibited an AUC of 86.56% for PD-1 expression and an AUC of 83.93% for PD-L1 expression, outperforming other deep learning and machine learning models.

CONCLUSIONS : We demonstrated that a non-invasive deep learning-based model trained with self-supervised contrastive learning could accurately predict the PD-1 and PD-L1 expression status, and might assist the precision treatment of patients withHCC, in particular the use of immune checkpoint inhibitors.

Xie Tianshu, Wei Yi, Xu Lifeng, Li Qian, Che Feng, Xu Qing, Cheng Xuan, Liu Minghui, Yang Meiyi, Wang Xiaomin, Zhang Feng, Song Bin, Liu Ming

2023

PD-1/L1, computed tomography, contrastive learning, deep learning, hepatocellular carcinoma, self-supervised learning

Surgery Surgery

Investigation of trends in gut microbiome associated with colorectal cancer using machine learning.

In Frontiers in oncology

BACKGROUND : The rapid growth of publications on the gut microbiome and colorectal cancer (CRC) makes it feasible for text mining and bibliometric analysis.

METHODS : Publications were retrieved from the Web of Science. Bioinformatics analysis was performed, and a machine learning-based Latent Dirichlet Allocation (LDA) model was used to identify the subfield research topics.

RESULTS : A total of 5,696 publications related to the gut microbiome and CRC were retrieved from the Web of Science Core Collection from 2000 to 2022. China and the USA were the most productive countries. The top 25 references, institutions, and authors with the strongest citation bursts were identified. Abstracts from all 5,696 publications were extracted for a text mining analysis that identified the top 50 topics in this field with increasing interest. The colitis animal model, expression of cytokines, microbiome sequencing and 16s, microbiome composition and dysbiosis, and cell growth inhibition were increasingly noticed during the last two years. The 50 most intensively investigated topics were identified and further categorized into four clusters, including "microbiome sequencing and tumor," "microbiome compositions, interactions, and treatment," "microbiome molecular features and mechanisms," and "microbiome and metabolism."

CONCLUSION : This bibliometric analysis explores the historical research tendencies in the gut microbiome and CRC and identifies specific topics of increasing interest. The developmental trajectory, along with the noticeable research topics characterized by this analysis, will contribute to the future direction of research in CRC and its clinical translation.

Yu Chaoran, Zhou Zhiyuan, Liu Bin, Yao Danhua, Huang Yuhua, Wang Pengfei, Li Yousheng

2023

Latent Dirichlet Allocation, Web of Science, bibliometric, colorectal cancer, microbiome

General General

Configural relations in humans and deep convolutional neural networks.

In Frontiers in artificial intelligence

Deep convolutional neural networks (DCNNs) have attracted considerable interest as useful devices and as possible windows into understanding perception and cognition in biological systems. In earlier work, we showed that DCNNs differ dramatically from human perceivers in that they have no sensitivity to global object shape. Here, we investigated whether those findings are symptomatic of broader limitations of DCNNs regarding the use of relations. We tested learning and generalization of DCNNs (AlexNet and ResNet-50) for several relations involving objects. One involved classifying two shapes in an otherwise empty field as same or different. Another involved enclosure. Every display contained a closed figure among contour noise fragments and one dot; correct responding depended on whether the dot was inside or outside the figure. The third relation we tested involved a classification that depended on which of two polygons had more sides. One polygon always contained a dot, and correct classification of each display depended on whether the polygon with the dot had a greater number of sides. We used DCNNs that had been trained on the ImageNet database, and we used both restricted and unrestricted transfer learning (connection weights at all layers could change with training). For the same-different experiment, there was little restricted transfer learning (82.2%). Generalization tests showed near chance performance for new shapes. Results for enclosure were at chance for restricted transfer learning and somewhat better for unrestricted (74%). Generalization with two new kinds of shapes showed reduced but above-chance performance (≈66%). Follow-up studies indicated that the networks did not access the enclosure relation in their responses. For the relation of more or fewer sides of polygons, DCNNs showed successful learning with polygons having 3-5 sides under unrestricted transfer learning, but showed chance performance in generalization tests with polygons having 6-10 sides. Experiments with human observers showed learning from relatively few examples of all of the relations tested and complete generalization of relational learning to new stimuli. These results using several different relations suggest that DCNNs have crucial limitations that derive from their lack of computations involving abstraction and relational processing of the sort that are fundamental in human perception.

Baker Nicholas, Garrigan Patrick, Phillips Austin, Kellman Philip J

2022

DCNNs, abstract relations, abstract representation, deep convolutional neural networks, deep learning, perception of relations, shape perception, visual relations