Cancer Detection Performance Research Articles

Automated lung cancer detection using novel genetic TPOT feature optimization with deep learning techniques

Lung cancer remains a leading cause of cancer-related deaths globally. Early and accurate detection is crucial for improving patient outcomes. Traditional methods, relying on manual interpretation of medical images, are time-consuming and prone to errors. Deep learning, particularly convolutional neural networks (CNNs), offers an automated alternative capable of learning intricate patterns from medical images. However, previous deep learning models for lung cancer detection have faced challenges such as limited data, inadequate feature extraction, interpretability issues, and susceptibility to data variability. This paper presents a novel deep learning methodology that addresses these limitations. Our approach leverages expansive datasets, incorporates advanced feature extraction techniques, improves interpretability, and accommodates the diverse nature of lung cancer. Specifically, we develop dedicated models for both chest X-ray and CT images utilizing publicly available datasets from Kaggle. Through the integration of feature selection and model selection techniques—such as employing a genetic algorithm in conjunction with the tree-based pipeline optimization tool (TPOT)—we achieved remarkable accuracy. Our X-ray model attains an overall accuracy of 95.47%, the CT model achieves an accuracy of 98.70%, and the combined model achieves an impressive overall accuracy of 98.93%. Our methodology significantly enhances the performance and efficiency of lung cancer detection and is a valuable tool for early diagnosis and intervention.

A Review on Lung and Colon Combine Cancer Detection using ML and DL Techniques

The detection of lung and colon cancer is a critical challenge in medical diagnosis, and machine learning (ML) and deep learning (DL) techniques are increasingly being used to enhance accuracy and efficiency. This review focuses on the integration of ML and DL methods for the combined detection of lung and colon cancer, emphasizing their strengths, limitations, and future potential. The motivation behind this study is to address the growing demand for accurate and early detection of these cancers, which significantly impacts treatment outcomes. Current methods often struggle with feature complexity, image variability, and computational intensity, which limit their real-world applicability. The aim is to consolidate various ML and DL techniques that have been employed for this purpose, highlighting how hybrid models can improve detection rates. The objective of this review is to provide a comprehensive analysis of different methodologies, their datasets, pre-processing techniques, feature extraction methods, and evaluation parameters. This review also explores recent advancements, such as transfer learning combined with fine-tuning techniques, which can further optimize performance in cancer detection. The findings suggest that while current methods show promise, further improvements in model generalization, interpretability, and computational efficiency are required to overcome existing limitations and expand clinical use.

ENHANCED TRANSFORMER-BASED DEEP KERNEL FUSED SELF ATTENTION MODEL FOR LUNG NODULE SEGMENTATION AND CLASSIFICATION

Lung cancer remains a leading cause of mortality globally, with outcomes heavily dependent on the timeliness and accuracy of diagnosis. Traditional medical imaging techniques, while foundational in detecting lung nodules, often falter in distinguishing malignant from benign lesions with high precision, largely due to their inability to contextualize the complex spatial relationships within the images. Precise segmentation and classification of lung nodules is crucial for the early detection of lung cancer. This paper presents a novel deep learning model that incorporates a transformer block to improve the performance of lung cancer detection and classification. From performance evaluation, it is evident that our proposed model has an average accuracy of 93%. 05%, which is superior to the existing D3DR _ MKCA model with a mean accuracy of 91.53%. These findings are especially important for the identification of Adenocarcinoma and Small Cell Carcinoma, as improvements in the precision and recall factors have been achieved in these cases.

Diagnostic Performance of Faecal Immunochemical Testing (FIT) in Patients with Lynch Syndrome Scheduled for Colonoscopic Surveillance.

Lynch syndrome (LS) carries a substantial lifetime risk of colorectal cancer which is currently mitigated by biennial colonoscopy surveillance. Paramount to the surveillance programme is the removal of adenomas before malignant transformation but there is an associated service burden and morbidity of repeated endoscopy. We investigated if faecal immunochemical testing (FIT) for faecal haemoglobin has the diagnostic performance to replace colonoscopy. In this retrospective cohort study, patients due to undergo planned surveillance for LS between November 2020 and April 2022 were sent two FIT kits prior to colonoscopy. Test diagnostic performance of colorectal cancer (CRC), advanced and non-advanced adenoma detection was calculated for single and double FIT strategies. A faecal-Hb of 10 µg Hb/g was considered positive. In total, 78 patients, with 45 (57.7%) female, median age 52 years (IQR 41-63), completed at least one FIT and colonoscopy. The median time from FIT to colonoscopy was 47 days. A single FIT was positive in 7/30 cases of adenoma (2/3 advanced, 5/27 non-advanced). A total of 64 (82.1% of FIT1T returners) completed a second FIT. Using the greatest of the two FITs (FIT2TMAX) 8/26 (2/3 advanced, 4/23 non-advanced), patients with adenomas were identified. There were no cases of CRC. The sensitivity for adenoma detection was 23.3% and 23.1%, respectively. In patients with LS awaiting colonoscopy, FIT has a low sensitivity for detecting adenomas and advanced adenomas. This is not improved by the addition of a second FIT test.

Breast Lesion Detection for Ultrasound Images Using MaskFormer.

This study evaluates the performance of the MaskFormer model for segmenting and classifying breast lesions using ultrasound images, addressing ultrasound's limitations. Ultrasound used for breast cancer detection faces challenges like low image contrast and difficulty in the detection of small or multiple lesions, further complicated by variability based on operator skill. Initial experiments with U-Net and other CNN-based models revealed constraints, such as early plateauing in model loss, which indicated suboptimal learning and performance. In contrast, MaskFormer demonstrated continuous improvement, achieving higher precision in breast lesion segmentation and significantly reducing both false positives and false negatives. Comparative analysis showed MaskFormer's superior performance, with the highest precision and recall rates for malignant lesions and an overall mean average precision (mAP) of 0.943. The model's ability to detect a diverse range of breast lesions, including those potentially missed by the human eye, especially by less experienced practitioners, underscores its potential. These findings suggest that integrating AI models like MaskFormer could greatly enhance ultrasound performance for breast cancer detection, providing reliable, operator-independent image analysis and potentially improving patient outcomes on a global scale.

Automated Breast Ultrasound With Remote Reading for Primary Breast Cancer Screening: A Prospective Study Involving 46 Community Health Centers in China.

Background: China has faced barriers to implementation of a population-based mammographic screening program. Breast ultrasound provides an alternative screening modality to mammography in low-resource settings. Objective: To evaluate the performance of ABUS with remote reading as the primary screening modality for breast cancer. Methods: This prospective study enrolled asymptomatic women 35-69 years old from 46 community health centers across 18 provinces representing all six regions of China from January 2021 to December 2021. Participants underwent screening ABUS as the sole breast cancer screening modality, with images acquired by a technologist at the community health center. The 3D volumetric data were transferred via cloud-based software to a single remote reading center, where examinations were interpreted independently in batches by two subspecialized breast radiologists using BI-RADS; a third radiologist at the remote reader center resolved discrepancies. Diagnostic reports were returned to the community centers, and patients sought follow-up care at local hospitals. The reference standard incorporated a combination of histopathology and 24-month follow-up. Outcomes measures included cancer detection rate, abnormal interpretation rate (AIR), sensitivity, specificity, biopsy rate, and PPV. Results: The final analysis included 5978 enrolled participants (median age, 46 years [IQR 40-52 years]) who underwent screening ABUS at the community health centers with subsequent remote reading. A total of 24 ABUS-detected cancers and two interval cancers were diagnosed. The cancer detection rate was 4.0 per 1000 women (95% CI: 2.7-5.9), and the AIR was 11.9% (95% CI: 11.1-12.7). A total of 95.8% (23/24) of ABUS-detected cancers were invasive. The 23 invasive cancers had a median diameter of 10.0 mm; 73.9% (17/23) were node-negative. Sensitivity was 92.3% (95% CI: 75.9-97.9), and specificity was 88.4% (95% CI: 87.6-89.2). The biopsy rate was 1.7% (95% CI: 1.4-2.0), and the PPV of biopsy was 24.0% (95% CI: 16.7-33.2). Conclusion: ABUS screening with remote reading met benchmark performance for cancer detection in comparison to mammography, with infrequent interval cancers. Clinical Impact: ABUS with remote reading holds promise in enhancing access to breast cancer screening and early detection in low-resource settings or underserved regions where mammographic screening is not established.

Bladder Cancer detection by urinary methylation markers GHSR/MAL: a validation study

PurposeAlthough cystoscopy is a reliable tool for detecting bladder cancer, it poses a high burden on patients and entails high costs. This highlights the need for non-invasive and cost-effective alternatives. This study aimed to validate a previously developed urinary methylation marker panel containing GHSR and MAL.MethodsWe enrolled 134 patients who underwent cystoscopy because of hematuria, including 63 individuals with primary bladder cancer and 71 with non-malignant findings. Urine samples were self-collected at home and sent via regular mail. Subsequently, DNA was extracted and the hypermethylation of GHSR and MAL was evaluated using quantitative methylation-specific polymerase chain reaction. The performance of methylation markers was assessed using area-under-the-curve (AUC) analysis and sensitivity and specificity based on pre-established cut-off values.ResultsValidation of the marker panel GHSR/MAL resulted in an AUC of 0.87 at 79% sensitivity and 80% specificity. Sensitivity was comparable to the previous investigation (P > 0.9), though specificity was significantly lower (P = 0.026). Sensitivity was higher for high-grade tumors compared to low-grade tumors (94% vs. 60%, P = 0.002).ConclusionValidation of the GHSR/MAL methylation marker panel on at home collected urine samples confirms its robust performance for bladder cancer detection in a hematuria population, and underscores the diagnostic potential for future clinical application.

Biomarkers suitable for early detection of intrathoracic cancers in primary care: a systematic review.

Intrathoracic cancers, including lung cancer, mesothelioma, and thymoma, present diagnostic challenges in primary care. Biomarkers could resolve some challenges. We synthesized evidence on biomarkers performance for intrathoracic cancer detection in low-prevalence settings. A search in EMBASE and MEDLINE included studies that recruited participants with suspected intrathoracic cancer and reported on at least one diagnostic measure for a validated, non-invasive biomarker. Studies were excluded if participants were recruited based on a pre-established diagnosis. Fifty-two studies were included, reporting on 108 individual biomarkers and panels. CEA, CYFRA 21.1, and VEGF were evaluated for lung cancer and mesothelioma. For lung cancer, CEA and CYFRA 21.1 were most studied, with AUCs of 0.48-0.90 and 0.48-0.83, respectively. Pro-GRP and NSE had the highest NPVs (98.2%, 96.9%), while Early-CDT and MSC panels showed NPVs of 99.3% and 99.0% in smokers. For mesothelioma, Fibrillin-3 and mesothelin plus osteopontin had AUCs of 0.93 and 0.91, respectively. Thymoma panels (Binding AcHR + StrAb) and (Binding AcHR + Modulating AcHR + StrAb) had 100% NPVs in myasthenia gravis patients. The review highlights the performance of some biomarkers. However, few were evaluated in low-prevalence settings. Further evaluation is necessary before implementing these biomarkers for intrathoracic cancers in primary care.

Skin Cancer Diagnosis Using VGG16 and Transfer Learning: Analyzing the Effects of Data Quality over Quantity on Model Efficiency

The recent increase in the prevalence of skin cancer, along with its significant impact on individuals’ lives, has garnered the attention of many researchers in the field of deep learning models, especially following the promising results observed using these models in the medical field. This study aimed to develop a system that can accurately diagnose one of three types of skin cancer: basal cell carcinoma (BCC), melanoma (MEL), and nevi (NV). Additionally, it emphasizes the importance of image quality, as many studies focus on the quantity of images used in deep learning. In this study, transfer learning was employed using the pre-trained VGG-16 model alongside a dataset sourced from Kaggle. Three models were trained while maintaining the same hyperparameters and script to ensure a fair comparison. However, the quantity of data used to train each model was varied to observe specific effects and to hypothesize about the importance of image quality in deep learning models within the medical field. The model with the highest validation score was selected for further testing using a separate test dataset, which the model had not seen before, to evaluate the model’s performance accurately. This work contributes to the existing body of research by demonstrating the critical role of image quality in enhancing diagnostic accuracy, providing a comprehensive evaluation of the VGG-16 model’s performance in skin cancer detection and offering insights that can guide future improvements in the field.

Improving Computer-aided Detection for Digital Breast Tomosynthesis by Incorporating Temporal Change.

Purpose To develop a deep learning algorithm that uses temporal information to improve the performance of a previously published framework of cancer lesion detection for digital breast tomosynthesis. Materials and Methods This retrospective study analyzed the current and the 1-year-prior Hologic digital breast tomosynthesis screening examinations from eight different institutions between 2016 and 2020. The dataset contained 973 cancer and 7123 noncancer cases. The front end of this algorithm was an existing deep learning framework that performed single-view lesion detection followed by ipsilateral view matching. For this study, PriorNet was implemented as a cascaded deep learning module that used the additional growth information to refine the final probability of malignancy. Data from seven of the eight sites were used for training and validation, while the eighth site was reserved for external testing. Model performance was evaluated using localization receiver operating characteristic curves. Results On the validation set, PriorNet showed an area under the receiver operating characteristic curve (AUC) of 0.931 (95% CI: 0.930, 0.931), which outperformed both baseline models using single-view detection (AUC, 0.892 [95% CI: 0.891, 0.892]; P < .001) and ipsilateral matching (AUC, 0.915 [95% CI: 0.914, 0.915]; P < .001). On the external test set, PriorNet achieved an AUC of 0.896 (95% CI: 0.885, 0.896), outperforming both baselines (AUC, 0.846 [95% CI: 0.846, 0.847]; P < .001 and AUC, 0.865 [95% CI: 0.865, 0.866]; P < .001, respectively). In the high sensitivity range of 0.9 to 1.0, the partial AUC of PriorNet was significantly higher (P < .001) relative to both baselines. Conclusion PriorNet using temporal information further improved the breast cancer detection performance of an existing digital breast tomosynthesis cancer detection framework. Keywords: Digital Breast Tomosynthesis, Computer-aided Detection, Breast Cancer, Deep Learning © RSNA, 2024 See also commentary by Lee in this issue.

Si Nanorod Array Integrated Microfluidic Device for Enhanced Extracellular Vesicle Isolation

AbstractTumor‐derived extracellular vesicles (EVs) have attracted tremendous interest as one of the early cancer diagnostic markers. The major obstacle preventing EV‐based liquid biopsy is the efficient collection of EVs from the complex body fluid environment. This paper introduces a nanorod‐integrated microfluidic chip capable of immunoaffinity‐isolating EVs. Periodic silicon nanorod arrays in zigzag channels are prepared by nanosphere lithography. Nanorod sidewalls provide larger binding sites for antibodies, and their close interspacing to the EV sizes improves the binding probability. The fluid simulation results show that the significant increase in isolation efficiency also comes from the liquid perturbation enhanced by the particular nanorod arrangement. Under optimal operating conditions, plasma samples from patients (n = 14) with different types of cancers (hepatocellular carcinoma, colorectal cancer, and pancreatic adenocarcinoma) to the chip for EV isolation is applied. In this proof‐of‐concept study, the expression level of the epidermal growth factor receptor (EGFR) in isolated EVs is then quantified using droplet digital PCR, showing good diagnostic performance in cancer detection.

AI-enhanced Mammography With Digital Breast Tomosynthesis for Breast Cancer Detection: Clinical Value and Comparison With Human Performance.

Purpose To compare two deep learning-based commercially available artificial intelligence (AI) systems for mammography with digital breast tomosynthesis (DBT) and benchmark them against the performance of radiologists. Materials and Methods This retrospective study included consecutive asymptomatic patients who underwent mammography with DBT (2019-2020). Two AI systems (Transpara 1.7.0 and ProFound AI 3.0) were used to evaluate the DBT examinations. The systems were compared using receiver operating characteristic (ROC) analysis to calculate the area under the ROC curve (AUC) for detecting malignancy overall and within subgroups based on mammographic breast density. Breast Imaging Reporting and Data System results obtained from standard-of-care human double-reading were compared against AI results with use of the DeLong test. Results Of 419 female patients (median age, 60 years [IQR, 52-70 years]) included, 58 had histologically proven breast cancer. The AUC was 0.86 (95% CI: 0.85, 0.91), 0.93 (95% CI: 0.90, 0.95), and 0.98 (95% CI: 0.96, 0.99) for Transpara, ProFound AI, and human double-reading, respectively. For Transpara, a rule-out criterion of score 7 or lower yielded 100% (95% CI: 94.2, 100.0) sensitivity and 60.9% (95% CI: 55.7, 66.0) specificity. The rule-in criterion of higher than score 9 yielded 96.6% sensitivity (95% CI: 88.1, 99.6) and 78.1% specificity (95% CI: 73.8, 82.5). For ProFound AI, a rule-out criterion of lower than score 51 yielded 100% sensitivity (95% CI: 93.8, 100) and 67.0% specificity (95% CI: 62.2, 72.1). The rule-in criterion of higher than score 69 yielded 93.1% (95% CI: 83.3, 98.1) sensitivity and 82.0% (95% CI: 77.9, 86.1) specificity. Conclusion Both AI systems showed high performance in breast cancer detection but lower performance compared with human double-reading. Keywords: Mammography, Breast, Oncology, Artificial Intelligence, Deep Learning, Digital Breast Tomosynthesis © RSNA, 2024.

Deep learning model integrating cfDNA methylation and fragment size profiles for lung cancer diagnosis

Detecting aberrant cell-free DNA (cfDNA) methylation is a promising strategy for lung cancer diagnosis. In this study, our aim is to identify methylation markers to distinguish patients with lung cancer from healthy individuals. Additionally, we sought to develop a deep learning model incorporating cfDNA methylation and fragment size profiles. To achieve this, we utilized methylation data collected from The Cancer Genome Atlas and Gene Expression Omnibus databases. Then we generated methylated DNA immunoprecipitation sequencing and genome-wide Enzymatic Methyl-seq (EM-seq) form lung cancer tissue and plasma. Using these data, we selected 366 methylation markers. A targeted EM-seq panel was designed using the selected markers, and 142 lung cancer and 56 healthy samples were produced with the panel. Additionally, cfDNA samples from healthy individuals and lung cancer patients were diluted to evaluate sensitivity. Its lung cancer detection performance reached an accuracy of 81.5% and an area under the receiver operating characteristic curve of 0.87. In the serial dilution experiment, we achieved tumor fraction detection of 1% at 98% specificity and 0.1% at 80% specificity. In conclusion, we successfully developed and validated a combination of methylation panel and a deep learning model that can distinguish between patients with lung cancer and healthy individuals.

Optimasi Convolutional Neural Networks untuk Deteksi Kanker Payudara menggunakan Arsitektur DenseNet

Breast cancer is a disease commonly suffered by women worldwide, ranking as the second-largest disease burden. In response to the urgent need for improved detection accuracy, Convolutional Neural Networks (CNNs) promise significant advancements. The objective of this research is to optimize the use of CNNs with the DenseNet architecture for breast cancer detection. The study employs quantitative methods, leveraging Deep Learning through CNNs. Mammography data is sourced from Kaggle, specifically the “Breast Histopathology Images” dataset. This dataset comprises 90,000 digital mammography images, which are preprocessed and divided proportionally for training, validation, and model testing. Research variables encompass CNN model parameters, training techniques, and the integration of imaging modalities to enhance breast cancer detection performance. The research focuses on processed mammography data, with accuracy and image quality as key evaluation metrics for breast cancer sample identification. Our findings demonstrate that the DenseNet architecture within CNNs achieves an impressive 92% accuracy in breast cancer detection. This remarkable performance signifies success in enhancing image quality and class prediction, aligning with the DenseNet architecture’s flow diagram. Ultimately, these results contribute significantly to effective breast cancer diagnosis by optimizing CNNs with the DenseNet architecture to improve image quality during breast cancer sampling.

Assessment of plasma cell-free DNA fragmentation for multi-cancer early detection: An independent clinical validation study.

e15037 Background: Cell-free DNA (cfDNA) in the circulation has gained significant attention due to its potential applications for non-invasive early cancer detection. Fragmentomics of cfDNA is emerging as a promising field of biomarker research, exhibiting the capability to sensitively detect multiple cancer types. In previous retrospective research, we developed an approach called PatternWGS for comprehensive analysis of cfDNA fragmentation patterns and presented a multi-cancer detection model. In this study, we aim to evaluate the performance of PatternWGS in an independent validation cohort including lung, breast, colorectal, liver, and gastric cancers from a new clinical center. Methods: Participants with confirmed diagnosis of lung, breast, colorectal or liver cancer were enrolled in the cancer group. Non-cancer participants without known presence of malignancies were recruited from the same clinical center. Plasma samples from patients with lung (n=47), breast (n=19), colorectal (n=14), liver (n=8) and gastric (n=6) cancers, along with 101 non-cancer individuals were collected in this study. cfDNA was isolated from plasma and stored within a month prior to the whole genome sequencing. The characteristics of cfDNA fragmentation integrating fragment size profiles, motif patterns, genomic coverage distributions were utilized to evaluate the performance for cancer signal detection. In addition, the performance of estimated cfDNA tumor fraction, using copy number aberrations (ichorCNA), was compared with our approach. Results: 94 patients with one of five types of cancers and 101 non-cancer controls were prospectively enrolled between July 2023 and November 2023. Among the cancer group, 64% were at early stages (Stage I: 48%, Stage II: 16%). A significantly higher proportion of shorter fragments was observed in cancer group, and the degree increased with increasing cancer stage. Overall sensitivity across cancer types and stages was 72.3%, and the specificity for non-cancer controls was 80%. As expected, the sensitivity of cancer detection is higher in late-stage patients compared to those in the early-stages (61% in stage I, 80% in stage II, 84.6% in stage III, and 91.7% in stage IV). The sensitivity of detecting lung (72% had stage I cancer), breast, colorectal, liver, and gastric cancers reached 57.4%, 84.2%, 85.7%, 87.5% and 100% respectively. The detection performance using cfDNA tumor fraction by ichorCNA demonstrated a relatively low overall sensitivity of 28.7% (24.6% for stage I-II, 44% for stage III-IV). Conclusions: This independent clinical validation study, which focused on exploring plasma cfDNA fragmentation, demonstrated ideal performance in early-stage cancer detection. The cfDNA fragmentation-based approach exhibited a superior sensitivity in detecting early cancer signals compared with copy number aberrations-based approach.

Genome-Scale Multimodal Analysis of Cell-Free DNA Whole-Methylome Sequencing for Noninvasive Esophageal Cancer Detection.

Simultaneous profiling of cell-free DNA (cfDNA) methylation and fragmentation features to improve the performance of cfDNA-based cancer detection is technically challenging. We developed a method to comprehensively analyze multimodal cfDNA genomic features for more sensitive esophageal squamous cell carcinoma (ESCC) detection. Enzymatic conversion-mediated whole-methylome sequencing was applied to plasma cfDNA samples extracted from 168 patients with ESCC and 251 noncancer controls. ESCC characteristic cfDNA methylation, fragmentation, and copy number signatures were analyzed both across the genome and at accessible cis-regulatory DNA elements. To distinguish ESCC from noncancer samples, a first-layer classifier was developed for each feature type, the prediction results of which were incorporated to construct the second-layer ensemble model. ESCC plasma genome displayed global hypomethylation, altered fragmentation size, and chromosomal copy number alteration. Methylation and fragmentation changes at cancer tissue-specific accessible cis-regulatory DNA elements were also observed in ESCC plasma. By integrating multimodal genomic features for ESCC detection, the ensemble model showed improved performance over individual modalities. In the training cohort with a specificity of 99.2%, the detection sensitivity was 81.0% for all stages and 70.0% for stage 0-II. Consistent performance was observed in the test cohort with a specificity of 98.4%, an all-stage sensitivity of 79.8%, and a stage 0-II sensitivity of 69.0%. The performance of the classifier was associated with the disease stage, irrespective of clinical covariates. This study comprehensively profiles the epigenomic landscape of ESCC plasma and provides a novel noninvasive and sensitive ESCC detection approach with genome-scale multimodal analysis.

Detection and characterization of pancreatic and biliary tract cancers using cell-free DNA fragmentomics

BackgroundPlasma cell-free DNA (cfDNA) fragmentomics has demonstrated significant differentiation power between cancer patients and healthy individuals, but little is known in pancreatic and biliary tract cancers. The aim of this study is to characterize the cfDNA fragmentomics in biliopancreatic cancers and develop an accurate method for cancer detection.MethodsOne hundred forty-seven patients with biliopancreatic cancers and 71 non-cancer volunteers were enrolled, including 55 patients with cholangiocarcinoma, 30 with gallbladder cancer, and 62 with pancreatic cancer. Low-coverage whole-genome sequencing (median coverage: 2.9 ×) was performed on plasma cfDNA. Three cfDNA fragmentomic features, including fragment size, end motif and nucleosome footprint, were subjected to construct a stacked machine learning model for cancer detection. Integration of carbohydrate antigen 19–9 (CA19-9) was explored to improve model performance.ResultsThe stacked model presented robust performance for cancer detection (area under curve (AUC) of 0.978 in the training cohort, and AUC of 0.941 in the validation cohort), and remained consistent even when using extremely low-coverage sequencing depth of 0.5 × (AUC: 0.905). Besides, our method could also help differentiate biliopancreatic cancer subtypes. By integrating the stacked model and CA19-9 to generate the final detection model, a high accuracy in distinguishing biliopancreatic cancers from non-cancer samples with an AUC of 0.995 was achieved.ConclusionsOur model demonstrated ultrasensitivity of plasma cfDNA fragementomics in detecting biliopancreatic cancers, fulfilling the unmet accuracy of widely-used serum biomarker CA19-9, and provided an affordable way for accurate noninvasive biliopancreatic cancer screening in clinical practice.

Lung and Colon Cancer Histopathological Image Classification Using 1D Convolutional Channel-based Attention Networks

Lung and Colon cancer are the leading diseases of death and disability in humans caused by a combination of genetic diseases and biochemical abnormalities. If these are diagnosed in their early stages, they can not be spread in organs and negatively impact human life. Many deep-learning networks have recently been proposed to detect and classify these malignancies. However, incorrect detection or misclassification of these fatal diseases can significantly affect an individual's health and well-being. This paper introduces a novel, cost-effective, and mobile-embedded architecture to diagnose and classify Lung squamous cell carcinomas and adenocarcinomas of the lung and colon from digital pathology images. Extensive experiment shows that our proposed modifications achieve 100% testing results for lung, colon, and lung-and-colon cancer detection. Our novel architecture takes around 0.65 million trainable parameters and around 6.4 million flops to achieve the best lung and colon cancer detection performance. Compared with the other results, our proposed architecture shows state-of-the-art performance.

Enhancing breast cancer segmentation and classification: An Ensemble Deep Convolutional Neural Network and U-net approach on ultrasound images

Breast cancer is a condition where the irregular growth of breast cells occurs uncontrollably, leading to the formation of tumors. It poses a significant threat to women’s lives globally, emphasizing the need for enhanced methods of detecting and categorizing the disease. In this work, we propose an Ensemble Deep Convolutional Neural Network (EDCNN) model that exhibits superior accuracy compared to several transfer learning models and the Vision Transformer model. Our EDCNN model integrates the strengths of the MobileNet and Xception models to improve its performance in breast cancer detection and classification. We employ various preprocessing techniques, including image resizing, data normalization, and data augmentation, to prepare the data for analysis. By following these measures, the formatting is optimized, and the model’s capacity to make generalizations is improved. We trained and evaluated our proposed EDCNN model using ultrasound images, a widely available modality for breast cancer imaging. The outcomes of our experiments illustrate that the EDCNN model attains an exceptional accuracy of 87.82% on Dataset 1 and 85.69% on Dataset 2, surpassing the performance of several well-known transfer learning models and the Vision Transformer model. Furthermore, an AUC value of 0.91 on Dataset 1 highlights the robustness and effectiveness of our proposed model. Moreover, we highlight the incorporation of the Grad-CAM Explainable Artificial Intelligence (XAI) technique to improve the interpretability and transparency of our proposed model. Additionally, we performed image segmentation using the U-Net segmentation technique on the input ultrasound images. This segmentation process allowed for the identification and isolation of specific regions of interest, facilitating a more comprehensive analysis of breast cancer characteristics. In conclusion, the study presents a creative approach to detecting and categorizing breast cancer, demonstrating the superior performance of the EDCNN model compared to well-established transfer learning models. Through advanced deep learning techniques and image segmentation, this study contributes to improving diagnosis and treatment outcomes in breast cancer.

A serial image analysis architecture with positron emission tomography using machine learning combined for the detection of lung cancer

Introduction and objectivesLung cancer is the second type of cancer with the second highest incidence rate and the first with the highest mortality rate in the world. Machine learning through the analysis of imaging tests such as positron emission tomography/computed tomography (PET/CT) has become a fundamental tool for the early and accurate detection of cancer. The objective of this study was to propose an image analysis architecture (PET/CT) ordered in phases through the application of ensemble or combined machine learning methods for the early detection of lung cancer by analyzing PET/CT images. Material and methodsA retrospective observational study was conducted utilizing a public dataset entitled “A large-scale CT and PET/CT dataset for lung cancer diagnosis.” Various imaging modalities, including CT, PET, and fused PET/CT images, were employed. The architecture or framework of this study comprised the following phases: 1. Image loading or collection, 2. Image selection, 3. Image transformation, and 4. Balancing the frequency distribution of image classes. Predictive models for lung cancer detection using PET/CT images included: a) the Stacking model, which used Random Forest and Support Vector Machine (SVM) as base models and complemented them with a logistic regression model, and b) the Boosting model, which employed the Adaptive Boosting (AdaBoost) model for comparison with the Stacking model. Quality metrics used for evaluation included accuracy, precision, recall, and F1-score. ResultsThis study showed a general performance of 94% with the Stacking method and a general performance of 77% with the Boosting method. ConclusionsThe Stacking method proved to be a model with high performance and quality for lung cancer detection when analyzing PET/CT images.