Classification of Colorectal Cancer using ResNet and EfficientNet Models

Colorectal cancer is a malignant tumor with the third and second highest incidence and mortality rates worldwide, and its mortality rate is increasing annually. Colorectal cancer evolves gradually over a long period of time. The evolving process from colorectal adenoma to colorectal cancer takes approximately 10-15 years, providing a sufficiently long "window period" for early screening and diagnosis of colorectal cancer. The recurrence and mortality rates can be controlled at a low level with an early intervention. Metaxas et al sum-marized existing screening methods and their applicable scope in a recent publication. Moreover, they provide suggestions on how to improve adherence. This editorial provides a commentary on their article and discuss the roles of different screening methods in the early screening of colorectal cancer.

Three-dimensional ordered DNA network constructed by a biomarker pair for accurate monitoring of colorectal cancer

Colorectal cancer (CRC) is an important public health problem, especially in Europe. Every year approximately 435,000 new cases are being diagnosed with colorectal cancer. Half of these people die, which makes this type of cancer a second-rated mortality cause in Europe. The main aim of this paper was to make a literature search related to colorectal cancer with a focus on the developed Western European countries and countries close to the Republic of North Macedonia. This paper also intended to provide a picture of the early screening for colorectal cancer as the most efficient method for prevention and early detection of colorectal cancer and its application in some developed countries. For achieving the objectives set out in this paper, a survey of the available literature (both electronic and print) as well as of the grey literature was made. A systematic search of the following databases was made: PubMed, European Commission, and Google Scholar. The inclusion criteria were studies conducted between 2008 and 2020, with an emphasis on the newest studies and those published in the neighboring countries. According to the WHO recommendations and practices in modern countries, colorectal cancer should be identified as a serious public health problem. This includes unclear cause of its occurrence, as well as all possible risk factors which make it almost possible to implement an adequate prevention program. The most acceptable sensitive test for implementation of the screening program for colorectal cancer is the utilization of the FOB test. It is of great importance to follow-up patients with a positive FOB test as well as to offer easy access to health services, i.e., screening program to the population.

Background: Early screening for colorectal cancer (CRC) has the potential to improve patient prognosis, but current screening methods are limited. In this prospective study, we aimed to evaluate the multigene (Septin9, SDC2, KCNQ5, and IKZF1) detection in patient plasma for CRC diagnosis. Methods: Overall, 67 participants were enrolled, including 31 patients with CRC, 17 patients with colorectal polyp, and 19 normal controls who underwent colonoscopy. Carcinoembryonic antigen (CEA) and Septin9, SDC2, KCNQ5, and IKZF1 methylation tests were performed. Sensitivity, specificity, and the area under the receiver operating characteristic (ROC) curve were used to evaluate the diagnostic value of each biomarker. The association between positive rates of methylated Septin9, SDC2, KCNQ5, and IKZF1 and the clinicopathological characteristics of CRC was also analyzed. Results: The positive rate of multigene methylation detection was 87.1% (27/31) in patients with CRC, which was higher than single indicators: CEA (51.61%, 16/31), Septin9 (41.94%, 13/31), SDC2 (41.94%, 13/31), KCNQ5 (58.06%, 18/31), and IKZF1 (32.26%, 10/31). In the colorectal polyp group, the rate of multigene methylation detection is 88.24% (15/17), which was also higher than single indicator: CEA (17.65%, 3/17), Septin9 (11.76%, 2/17), SDC2 (64.71%, 11/17), KCNQ5 (58.82%, 10/17), and IKZF1 (35.29%, 6/17). The ROC curves further showed better diagnostic value of the multigene test for CRC than any single gene. Correlation analysis found that the positive rate of the test was not affected by patients' clinicopathologic characteristics. Conclusion: The combination of methylated Septin9, SDC2, KCNQ5, and IKZF1 tests is preferable to individual gene tests for patients with CRC and polyp.

Colorectal cancer (CRC) is a major public health threat. Therefore, CRC screening uptake has been a focus with the established precancerous lesion and the strong association of early detection with staging and survival of the disease. However, CRC screening is relatively low in many countries. This article briefly discussed the current situation of CRC, recommendations, and current uptake of CRC screening in various countries. Besides that, this article also highlights the potential factors that help to predict the CRC screening uptake worldwide. Identification of those factors could guide policymakers to develop an effective strategy to improve the CRC screening uptake and ultimately improve the health outcome of the population.Significance for public healthThis study highlights the public health challenge in early screening for colorectal cancer (CRC). The prevalence of never screened for faecal occult blood test is relatively high. This review dissects the issue and further discuss on the predictors, which could guide policymakers in developing strategy to improve CRC screening uptake.

BackgroundCurrently, due to the invasive nature of colonoscopy and the associated pain, people avoid undergoing the procedure, making it difficult to detect the majority of potential early stage colorectal carcinoma/precancerous lesions or advanced adenoma. Advanced colorectal adenoma is the main precursor to the development of colorectal carcinoma. Therefore, improving advanced colorectal adenoma detection rate can significantly decrease the development and morbidity of colorectal carcinoma. Accordingly, a non-invasive method to screen high-risk people for colonoscopy in clinical practice is urgently needed.Main textWith the development of medical technology, screening methods for colorectal carcinoma are emerging rapidly, and diverse non-invasive methods are being developed. Cell-free DNA (cfDNA), commonly referred to as liquid biopsy, has promising application prospects as a minimally invasive strategy for early screening of colorectal cancer. CfDNA has already been applied in the field of prenatal diagnosis, advanced carcinoma, and organ transplantation, and the application cfDNA in advanced colorectal adenoma is at the cutting-edge of current research. Thus, this review summarizes the progress in research on different biological characteristics of cfDNA and its utility in the screening of advanced colorectal adenoma, including sizes of cfDNA molecules, end signature of cfDNA (preferred ends, end motifs, jagged ends), nucleosomal footprints, cfDNA topology, cfDNA methylation, and cfDNA integrity.ConclusionsWe hope that this review will advance this promising research field.

BackgroundCertain “star intestinal bacteria” have been found to act as a contributor to the development of colorectal cancer (CRC). Besides, given that the gut microbiome can be detected in a diverse range of samples (stool, tissue, blood, etc), it is categorized into fecal microbiome, blood microbiome, and tissue microbiome.MethodsTo provide an overview of the recent research progress, this review summarizes the characteristics of the gut microbiome in different samples at each stage of CRC and their screening efficiency.ResultsThe screening models constructed from different sample microbiomes (healthy/colorectal adenoma, healthy/CRC, and colorectal adenoma/CRC) have both strengths and constraints in terms of biomarker reproducibility and area under the receiver‐operating characteristic curve (AUC) of the screening models. Many bacteria, such as Bifidobacteria, Fusobacterium nucleatum (F. n), Geotrichum candidum, Porphyromonas asaccharolytica, Escherichia coli, Rhodococcus, Anaerostipes caccae, Enhydrobacter, Lachnoclostridiumsp. m3, Bacteroides clarus, Clostridium hathewayi, Ruminococcaceae, Bacteroides thetaiotaomicron, Culinariside, and enterotoxigenic Bacteroides fragilis (ETBF), show favorable diagnostic efficacy in early screening of colorectal cancer.ConclusionsThis review highlights stool, blood, tissue, and bowel fluid are the main sample sources for biomarkers, each with its own advantages and limitations. Moreover, other samples such as extracellular vesicles and biofilms also should been deserved further attention.

Prostate cancer is the most prevalent solid tumor in males and one of the most common causes of male mortality. It is the most common type of cancer in men, a major global public health issue, and accounts for up to 7.3% of all male cancer diagnoses worldwide. To optimize patient outcomes and ensure therapeutic success, an accurate diagnosis must be made promptly. To achieve this, we focused on using ResNet50, a convolutional neural network (CNN) architecture, to analyze prostate histological images to classify prostate cancer. ResNet50, due to its efficiency in medical image classification, was used to classify the histological images as benign or malignant. In this study, a total of 1276 prostate biopsy images were used on the ResNet50 model. We employed evaluation metrics such as accuracy, precision, recall, and F1 score. The results showed that the ResNet50 model performed excellently with an overall accuracy of 0.98, 1.00 as precision, 0.98 as recall, and 0.97 as F1 score for benign. The malignant histological image has 0.99, 0.98, and 0.97 as precision, recall, and F1 scores. It also recorded a 95% confidence interval (CI) for accuracy as (0.91, 1.00) and a performance gain of 4.26% compared to MobileNet and CNN-RNN. The result of our model was also compared with the state-of-the-art (SOTA) DL models to ensure robustness. This study has demonstrated the potential of the ResNet50 model in the classification of prostate cancer. Again, the clinical integration of the results of this study will aid decision-makers in enhancing patient outcomes.

Colorectal cancer (CRC) is one of the most prevalent cancer types, but until now only few studies have addressed the problem of automatic recognition of distinct tissue types in histological CRC images. In this paper, we employ two deep convolutional neural networks, the original 50-layer residual network ResNet-50 and a modified residual network ResNet-50-fla-drop, to automatically perform eight-class tissue separation in histological CRC images. Experimental results showed that the proposed modified residual network ResNet-50-fla-drop can achieve a classification accuracy of 94.4% for eight-class CRC tissue separation, and it outperforms other published machine learning methods, including the radial-basis function SVM and VGG-VD-16.

Colorectal cancer (CRC) is currently the second deadliest cancer in the world. Apart from the known prognostic factors, the role of tumour microenvironment (TME), in particular the extracellular matrix (ECM) has gained traction in the past few years. This thesis focuses on characterisation of ECM features in CRC tissues by non-invasive and label-free imaging modalities – two-photon laser scanning microscopy (2PLSM) and matrix-assisted laser desorption ionisation mass spectrometry imaging (MALDI MSI). In manuscript I, by performing texture analysis on the second harmonic generation (SHG) signal emitted by collagen during 2PLSM, I identified differences in the morphological features of collagen in LSCC and RSCC (n =17). CRC tissues were found to contain more collagen fibres and a straighter morphology of collagen fibres in comparison to healthy colorectal tissues. The amount of collagen fibres was higher in LSCC compared to RSCC. However, these collagen fibres were found to be more chaotic in their organisation in RSCC compared to LSCC. There was no significant distinction in the waviness of the collagen fibres based on the anatomical origin of the tumour. Therefore, our findings suggest that while a single feature might not fully define collagen morphology, using a combination of features allows us to clearly and objectively outline the morphological characteristics. In manuscript II, I quantified peptide features in the abovementioned CRC patient cohort and combined them with imaging features to extract peptide signatures from distinct tumor regions. On performing receiver operator characteristics (ROC) analysis on 156 peptide signatures, I identified 46 peptides that were discriminative in tissues originating from LSCC compared to RSCC. I performed unsupervised, supervised and protein-protein interaction studies on these differential peaks. Using the unsupervised approach of principal component analysis, I depicted that the expression of principal component one is higher in LSCC than in RSCC. The supervised approach of k-means clustering led us to defining proteomic clusters in CRC tissues. Quantification of these clusters highlighted how the peptide clusters are differentially distributed in tumour tissues based on the anatomical origin of the primary tumour, i.e., LSCC or RSCC. viii The protein-protein interaction studies using STRING analysis determined that the differential proteins are functionally related. Major peptides upregulated in LSCC tissues were plectin (PCN), collagen (COL14A1, COL1A2, COL1A1, COL6A3), laminin (LAMA5, LAMB1), myosin (MYH11, MYH14), sorbin and SH3 domain-containing protein (SORBS1), endoplasmin (HSP90B1), transitional endoplasmic reticulum ATPase (TER ATPase), gelsolin (GSN), palladin (PALLD), and vinculin (VCL). These proteins collectively perform cellular functions such as focal adhesions, ECM assembly, cytoskeletal organization and cell migration. Furthermore, I performed correlative multimodal imaging to combine peptide features obtained from MALDI MSI with imaging features from 2PLSM and histology. I conducted 2PLSM experiments, followed by MALDI MSI and then histology on the same tissue section. I segmented nuclei from histology images and generated nuclei distribution heatmaps. I also performed texture analysis on SHG images to generate heatmaps of coherence in collagen fibre architecture. The heatmaps generated from histology and 2PLSM were used to define and annotate regions of low and high nuclei distribution, as well as low (chaotic) and high (organised) coherence in collagen fibres. On performing ROC analysis, I identified 55 discriminative peptides in the high nuclei distribution regions of LSCC in comparison to RSCC, such as x-ray cross-complementing protein- 5 (XRCC5) and Non-POU domain-containing octamer binding peptide (NONO). Both peptides have been previously correlated to poor prognosis in CRC. I further identified 45 peptides to be discriminative in the chaotic regions of tumours and 33 peptides which are differentially expressed in the organised regions of LSCC in comparison to RSCC. Notably, these peptides (HSP90B1, HSP90AB1, COL6A3, COL1A1, COL1A2, LAMA5, and LAMB1) are involved in the P13K-Akt signalling pathway, which plays a role in ECM mediated, epithelial-to mesenchymal transition (EMT) of malignant cells. In manuscript III, I identified imaging features from the collagen structure in tumours of a large patient cohort using convolutional neural network (CNN), a deep-learning method, to correlate these features to the recurrence of the tumour. I present a retrospective study on tissue sections obtained from a large cohort of 100 CRC patients. I optimized and performed 2PLSM on these tissues. Using the clinical information on the patients from the biobank database, I segregated the patients into recurrence and non-recurrence groups. I predicted tumourrecurrence in CRC patients ix with an accuracy of 84.2%, a precision of 87.0% and sensitivity/recall of 81.6%. I report that wavier and bundled streaks of collagen fibres in the CRC tissues are associated with tumour recurrence. Moreover, a pattern of collagen fibres surrounding the crypts is equally essential for identification of tumour recurrence. Alternatively, more straight, scattered and non-bundled collagen fibres are associated with non-recurrence. These findings about the differences in collagen morphology and the peptide signatures in the LSCC and RSCC tissues and the correlation of the collagen features with tumour recurrence in CRC tissues reveal novel insights on the ECM in CRC and shed light on the complexity and heterogeneity of the TME. In the future, these quantified 2PLSM metrics can be integrated into real-time imaging in endoscopes. The peptide signatures obtained in this study can help in identifying selective drug target targets for LSCC and RSCC. Furthermore, the correlative imaging method of applying 2PLSM with MALDI MSI and histology can be incorporated in other studies where the collagen structure and the peptide signatures are equally essential, such as connective tissue disorders. Lastly, the neural network that I trained to predict tumour recurrence on 2PLSM images can also be integrated into real-time prediction of tumour relapse via 2PLSM endoscopes. Thus, these findings contribute to a clearer understanding of CRC progression, metastasis and treatment strategies.

The exponential increase in Medical data and Computer-aided diagnostic tools has made it easier for Machine Learning (ML) Researchers to extract valuable insights from medical images resulting in better patient outcomes during Medical Summarization. According to current statistics, 8.8 million people died every year due to cancer death. It is therefore imperative to use these available image datasets for better diagnosis, increased patient outcomes and recovery from deadly diseases. As a result, this study proposes a Colorectal Cancer classification model for detecting KRAS mutation status of Patients through their Radiomic images using the transfer learning of a deep learning pipeline on Apache Spark. The datasets will be curated from National Cancer Institute (NCI), features extraction or classification of Computed tomography(CT) images into CT- based handcrafted radiomic signatures with a Convolutional Neural Network(CNN) using the Apache Spark framework, the system uploads the segmented scanning images to the High Distributed File System (HDFS), Confidence Interval(CI), Area Under Curve(AUC), Precision and Recall for Validation cohort, Docker for app deployment and containerization of the model developed for reproducibility, transfer learning and model reuse. The result would be a real-time model using Deep learning pipelines with Apache Spark and Keras Tensor flow for KRAS Mutation detection using Colorectal Cancer images. In conclusion, this research project would produce a scalable and reproducible model for the faster diagnosis of Colorectal Cancer through the KRAS mutation status of CRC Patients. Citation Format: Mary Adetutu Adewunmi. Scalable colorectal cancer (CRC) diagnosis model with arterial and unenhanced phases of KRAS mutational status using Apache Spark [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 6391.

Colorectal cancer become a significant public health issue and is the world’s second leading cause of death from cancer. Cancer becomes a very dangerous disease, because it gives no visible signs at an early stage. Signs of cancer will usually only be seen if it is in the third stadium or the last stadium, where the cancer has spread to surrounding organs. Early diagnosis of colorectal cancer is highly needed because treatment choices are decided and the period of survival is heavily affected. This paper proposes the Convolutional Neural Network (CNN) for detecting four classes of colon cancer. The data-set consists of 2500 images, divided into Tumor, Complex, Lymphoma and Stroma. This data set represents a selection from the Institute of Pathology, University of Heidelberg, Germany, consist of 150 × 150 px textures in histological pictures. The proposed system consist of two hidden convolutional layers, a fully connected layer and use Adam Optimizer with learning 0.001, and trained 10 times (epochs = 10). The result of the proposed system is 83% accuracy.

Early screening for colorectal cancer can detect early colorectal cancer, curb its further development, and extend the life expectancy. Colorectal cancer screening methods mainly include stool examination, imaging examination, endoscopic examination and the latest gene detection technology. Different screening methods have different effects. Stool examination is one of the most popular screening methods for the public. Key words: Colorectal neoplasms; Mass screening

Colorectal cancer (CRC) grading is typically carried out by assessing the degree of gland formation within histology images. To do this, it is important to consider the overall tissue micro-environment by assessing the cell-level information along with the morphology of the gland. However, current automated methods for CRC grading typically utilise small image patches and therefore fail to incorporate the entire tissue micro-architecture for grading purposes. To overcome the challenges of CRC grading, we present a novel cell-graph convolutional neural network (CGC-Net) that converts each large histology image into a graph, where each node is represented by a nucleus within the original image and cellular interactions are denoted as edges between these nodes according to node similarity. The CGC-Net utilises nuclear appearance features in addition to the spatial location of nodes to further boost the performance of the algorithm. To enable nodes to fuse multi-scale information, we introduce Adaptive GraphSage, which is a graph convolution technique that combines multi-level features in a data-driven way. Furthermore, to deal with redundancy in the graph, we propose a sampling technique that removes nodes in areas of dense nuclear activity. We show that modeling the image as a graph enables us to effectively consider a much larger image (around 16$\times$ larger) than traditional patch-based approaches and model the complex structure of the tissue micro-environment. We construct cell graphs with an average of over 3,000 nodes on a large CRC histology image dataset and report state-of-the-art results as compared to recent patch-based as well as contextual patch-based techniques, demonstrating the effectiveness of our method.

Area under the curve may hide poor generalisation to external datasets