Cancer Imaging Research Articles

Evaluation of functional magnetic resonance APT and DKI imaging for breast cancer

ObjectiveThis study aimed to compare the performance of amide proton transfer–weighted imaging (APTWI) and diffusion kurtosis imaging (DKI) in differentiating benign from malignant breast lesions, evaluate molecular subtypes of breast cancer, and determine the diagnostic efficacy of the quantitative magnetic resonance imaging (qMRI) parameters in differentiating benign from malignant breast diseases.MethodsThe study included 168 women who underwent breast APTWI and DKI at Yunnan Cancer Hospital between December 2022 and July 2023. The APT signal intensity (SI), apparent kurtosis coefficient (Kapp), non-Gaussian diffusion coefficient (Dapp), and apparent diffusion coefficient (ADC) values were measured before surgery. The differences in the aforementioned qMRI parameters in molecular subtypes of breast cancer were analyzed using one-way analysis of variance. The efficacy of each quantitative parameter in differentiating benign from malignant breast diseases was evaluated using the receiver-operating characteristic curve.ResultsSignificant differences in qMRI parameters were noted between benign and malignant breast lesions. The Kapp (P < .0001) and APT (P < .05) values were higher for malignant tumors than for benign lesions. Conversely, the ADC (P < .0001) and Dapp (P < .0001) values were lower for malignant tumors than for benign lesions. The diagnostic performance was assessed using the area under the curve (AUC) for various parameter combinations. The AUC of Kapp was 0.871, Dapp was 0.872, APT SI was 0.643, DKI + APT was 0.893, DKI + ADC was 0.936, APT + ADC was 0.925, and DKI + APT + ADC was 0.933. Additionally, ADC values (P = .01) demonstrated superior diagnostic performance compared to Kapp (P = .03), Dapp (P = .03), and APT values (P = .06) in distinguishing between different molecular subtypes of breast cancer.ConclusionsAPTWI distinguished benign from malignant breast disease and enhanced the utility of diffusion-weighted MRI. However, it was not superior to DKI and DWI in identifying the molecular subtypes of breast cancer.

Developmental trends and knowledge frameworks in the application of radiomics in prostate cancer: a bibliometric analysis from 2000 to 2024

BackgroundThis research utilized the bibliometrics method to analyze the published literature related to prostate cancer (PCa) imaging. Furthermore, current knowledge and research hotspots of radiomics in PCa diagnosis and treatment were comprehensively reviewed, as well as progress and emerging trends in field were explored.MethodsIn this investigation, the relevant literature on radiomics, and PCa was retrieved from Web of Science Core Collection (WoSCC) databases from 2000 and 2024. Furthermore, a comprehensive bibliometric analysis was carried out using advanced tools like CiteSpace6.2, VOS viewer, and the 'bibliometrix' package of R software to visualize the annual distribution of publications across various aspects such as authors, countries, journals, institutions, and keywords.ResultsThis analysis included 593 from 58 countries including China and the United States. Chinese Academy of Sciences and Frontiers in Oncology were the institutions and journals that publish the most relevant articles, -while Radiology journal had the greatest number of co-cited publications. Furthermore, 3,621 authors published on this topic, of which Madabhushi Anant and Stoyanova Radka had the highest contributions. Moreover, Lambin, P. had the most co-citations. In addition, the diagnostic characteristics of radiomics in PCa imaging and treatment strategies are the current research focal points. The establishment of multi-functional imaging techniques and independent factor models warrants future investigation.ConclusionsIn summary, this analysis revealed that the research on PCa imaging is developing vigorously, focusing on the diagnostic methods and intervention measures of imaging in PCa diagnosis and treatment. In the future, there is an urgent need for improved collaboration and communication among countries and institutions.

Realization of high-end PET devices that assist conventional PET devices in improving image quality via diffusion modeling

PurposeThis study aimed to implement high-end positron emission tomography (PET) equipment to assist conventional PET equipment in improving image quality via a distribution learning-based diffusion model.MethodsA diffusion model was first trained on a dataset of high-quality (HQ) images acquired by a high-end PET device (uEXPLORER scanner), and the quality of the conventional PET images was later improved on the basis of this trained model built on null-space constraints. Data from 180 patients were used in this study. Among them, 137 patients who underwent total-body PET/computed tomography scans via a uEXPLORER scanner at the Sun Yat-sen University Cancer Center were retrospectively enrolled. The datasets of 50 of these patients were used to train the diffusion model. The remaining 87 cases and 43 PET images acquired from The Cancer Imaging Archive were used to quantitatively and qualitatively evaluate the proposed method. The nonlocal means (NLM) method, UNet and a generative adversarial network (GAN) were used as reference methods.ResultsThe incorporation of HQ imaging priors derived from high-end devices into the diffusion model through network training can enable the sharing of information between scanners, thereby pushing the limits of conventional scanners and improving their imaging quality. The quantitative results showed that the diffusion model based on null-space constraints produced better and more stable results than those of the methods based on NLM, UNet and the GAN and is well suited for cross-center and cross-device imaging.ConclusionA diffusion model based on null-space constraints is a flexible framework that can effectively utilize the prior information provided by high-end scanners to improve the image quality of conventional scanners in cross-center and cross-device scenarios.

Intelligent Lung Cancer Detection: A Genetic Algorithm Machine Learning Fusion

Early detection of lung cancer is crucial for improving patient outcomes. While Deep Learning (DL) models have shown high accuracy in lung cancer diagnosis, they often require substantial computational resources. This study proposes a novel approach that leverages Genetic Algorithm (GA) to optimize feature selection and dimensionality reduction from lung cancer images. By integrating GA with conventional Machine Learning (ML) models, we demonstrate improved classification accuracy while minimizing computational requirements. Our experimental results show that combining GA with a feed-forward neural network classifier yields exceptional performance, achieving a classification accuracy of 99.70%. This approach offers a promising alternative to DL models for lung cancer detection, particularly in resource-constrained settings.

Breast cancer image classification by using HCNN and LeNet5

Medical data from many sectors has greatly increased during the last 10 years. One of the main causes of death and illness among women worldwide is breast cancer. Breast cancer (BC) is one of the most common cancers in women; the death rate is high and holds the second position next to lung cancer. Breast cancer develops when cells in the mammary glands and the ducts that transfer milk to the nipple grow out of control. This is the first step in the progression of the disease. However, the time complexity of the available techniques is immense due to the many processes. Additionally, the current research attempts to improve the computation time involved in the detection process. Therefore, an effective hybrid deep learning model is introduced to improve the prediction performance and reduce the time consumption compared to the machine learning model. The breast cancer dataset, obtained from Kaggle, is used as the input data. A Wiener filter preprocessing technique is applied to enhance the image quality, with an active Wiener filter employed for this purpose. The segmentation step is achieved using a Modified Watershed Algorithm, which isolates the region of interest within the images. Finally, classification is performed using a hybrid deep learning model. This model combines a Convolutional Neural Network (CNN) with an Enhanced Recurrent Neural Network (ERNN), leveraging the strengths of both architectures. According to experimental results, the proposed Hybrid Convolutional Neural Network (HCNN) model achieves an accuracy of 96.12%, a precision of 96.99%, a recall of 97.52%, and an F-measure of 97.25%, outperforming other existing models.

A Novel Approach for Lung Cancer Segmentation and Classification Using MSF-Customized ResNet152

This study presents a novel Multi-Scale Fusion-based Customized ResNet152 (MSF-Customized ResNet152) model for lung cancer segmentation and classification, designed to address challenges in class imbalance and feature extraction. The model combines multi-scale feature fusion with depth-wise separable convolution to capture fine and broad image details, enhancing segmentation accuracy. To address the minority representation of cancerous regions, a weighted recall loss was integrated with cross-entropy loss, optimized via Grid Search, to increase sensitivity. Additionally, data augmentation techniques were applied to expand the cancerous image set, further improving detection robustness. The model was evaluated on two datasets: a chest CT-scan dataset and a lung and colon cancer histopathology dataset. Using a 5-fold cross-validation approach, the MSF-Customized ResNet152 achieved a classification accuracy of 98.5%, precision of 98.2%, specificity of 97.2%, F1-score of 97.4%, and Dice coefficient of 97.3%. Comparative analysis demonstrated the superiority of the combined recall and cross-entropy loss function over standard methods, validating the model's effectiveness in handling class imbalance.

Distribution-on-scalar Single-index Quantile Regression Model for Handling Tumor Heterogeneity

This paper develops a distribution-on-scalar single-index quantile regression modeling framework to investigate the relationship between cancer imaging responses and scalar covariates of interest while tackling tumor heterogeneity. Conventional association analysis methods typically assume that the imaging responses are well-aligned after some preprocessing steps. However, this assumption is often violated in practice due to imaging heterogeneity. Although some distribution-based approaches are developed to deal with this heterogeneity, major challenges have been posted due to the nonlinear subspace formed by the distributional responses, the unknown nonlinear association structure, and the lack of statistical inference. Our method can successfully address all the challenges. We establish both estimation and inference procedures for the unknown functions in our model. The asymptotic properties of both estimation and inference procedures are systematically investigated. The finite-sample performance of our proposed method is assessed by using both Monte Carlo simulations and a real data example on brain cancer images from TCIA-GBM collection.

A Prostate Imaging-Reporting and Data System version 2.1-based predictive model for clinically significant prostate cancer diagnosis.

To develop and validate a Prostate Imaging-Reporting and Data System (PI-RADS) version 2.1 (v2.1)-based predictive model for diagnosis of clinically significant prostate cancer (csPCa), integrating clinical and multiparametric magnetic resonance imaging (mpMRI) data, and compare its performance with existing models. We retrospectively analysed data from patients who underwent prospective mpMRI assessment using the PI-RADS v2.1 scoring system and biopsy at our institution between April 2019 and December 2023. A 'Clinical Baseline' model using patient demographics and laboratory results and an 'MRI Added' model additionally incorporating PI-RADS v2.1 scores and prostate volumes were created and validated on internal and external patients. Both models were compared against two previously published MRI-based algorithms for csPCa using area under the receiver operating characteristic curve (AUC) and decision curve analysis. A total of 1319 patients across internal and external cohorts were included. Our 'MRI Added' model demonstrated significantly improved discriminative ability (AUCinternal 0.88, AUCexternal 0.79) compared to our 'Clinical Baseline' model (AUCinternal 0.75, AUCexternal 0.68) (P < 0.001). The 'MRI Added' model also showed higher net benefits across various clinical threshold probabilities and compared to a 'biopsy all' approach, it reduced unnecessary biopsies (defined as biopsies without Gleason Grade Group ≥2 csPCa) by 27% in the internal cohort and 10% in the external cohort at a risk threshold of 25%. However, there was no significant difference in predictive ability and reduction in unnecessary biopsies between our model and comparative ones developed for PI-RADS v2 and v1. Our PI-RADS v2.1-based mpMRI model significantly enhances csPCa prediction, outperforming the traditional clinical model in accuracy and reduction of unnecessary biopsies. It proves promising across diverse patient populations, establishing an updated, integrated approach for detection and management of prostate cancer.

Bounded Multivariate Contaminated Normal Mixture Model with an Application in Skin Cancer Detection

Introduction: In real-world datasets, outliers are a common occurrence that can have a significant impact on the accuracy and reliability of statistical analyses. Detecting these outliers and developing robust models to handle their presence is a crucial challenge in data analysis. For instance, natural images may have complex distributions of values due to environmental factors like noise and illumination, resulting in objects with overlapping regions and non-trivial contours that cannot be accurately described by Gaussian mixture models. In many real life applications, observed data always fall in bounded support regions. This leads to the idea of bounded support mixture models. Motivated by the aforementioned observations, we introduce a bounded multivariate cntaminated normal distribution for fitting data with non-Gaussian distributions, asymmetry, and bounded support which makes finite mixture models more robust to fitting, since rare observations are given less importance in calculations. Methods: A family of finite mixtures of bounded multivariate contaminated normal distributions is introduced. The model is well-suited for computer vision and pattern recognition problems due to its heavily-tailed and bounded nature, providing flexibility in modeling data in the presence of outliers. A feasible expectation- maximization algorithm is developed to compute the maximum likelihood estimates of the model parameters using a selection mechanism. Results: The proposed methodology is validated by conducting experiments on both simulated data and two real natural skin cancer images. We estimate the parameters by the proposed expectation-maximization algorithm. The obtained results shown that the proposed model has successfully enhanced accuracy in segmenting skin lesions. Conclusion: The reliable model-based clustering using finite mixtures of bounded multivariate contaminated normal distributions is introduced. An expectation-maximization algorithm was created to estimate parameters, with closed-form expressions utilized at the E-step. Practical tests on images for skin cancer detection showed enhanced accuracy in delineating skin lesions.

Importance of parametric and molecular imaging for therapeutic management of breast cancer

In recent years, various magnetic resonance (MRI) and positron emission tomography (PET) parameters have been investigated in breast cancer. Parametric imaging focuses on the visualization and quantification of biological, physiological, and pathological processes at the cellular and molecular level. It therefore provides important insights into the key processes in carcinogenesis and tumor progression. This article aims to illustrate the importance for the management of breast cancer. Based on the current literature, an overview of the current state of parametric breast imaging and its importance in therapy management is given. Moreover, future opportunities and challenges are highlighted. Parametric imaging in breast cancer includes MRI, nuclear medicine procedures such as PET, the combination of different techniques (PET-CT, PET-MRI) and the use of specific radiotracers. Parametric MRI of the breast mainly uses T2 and diffusion-weighted imaging (DWI) as well as dynamic contrast-enhanced MRI (CM-MRI). Quantitative and qualitative imaging biomarkers provide insights into tumor biology and allow conclusions to be drawn about the molecular subtype or prognosis. Recently, parametric imaging has become established in breast diagnostics. It is constantly evolving and will continue to gain in importance in the forthcoming years. It offers the opportunity to improve the diagnosis and treatment management of breast cancer.

Self-supervised learning improves robustness of deep learning lung tumor segmentation models to CT imaging differences.

Self-supervised learning (SSL) is an approach to extract useful feature representations from unlabeled data, and enable fine-tuning on downstream tasks with limited labeled examples. Self-pretraining is a SSL approach that uses curated downstream task dataset for both pretraining and fine-tuning. Availability of large, diverse, and uncurated public medical image sets presents the opportunity to potentially create foundation models by applying SSL in the "wild" that are robust to imaging variations. However, the benefit of wild- versus self-pretraining has not been studied for medical imageanalysis. Compare robustness of wild versus self-pretrained models created using convolutional neural network (CNN) and transformer (vision transformer [ViT] and hierarchical shifted window [Swin]) models for non-small cell lung cancer (NSCLC) segmentation from 3D computed tomography (CT) scans. CNN, ViT, and Swin models were wild-pretrained using unlabeled 10,412 3D CTs sourced from the cancer imaging archive and internal datasets. Self-pretraining was applied to same networks using a curated public downstream task dataset (n = 377) of patients with NSCLC. Pretext tasks introduced in self-distilled masked image transformer were used for both pretraining approaches. All models were fine-tuned to segment NSCLC (n = 377 training dataset) and tested on two separate datasets containing early (public n = 156) and advanced stage (internal n = 196) NSCLC. Models were evaluated in terms of: (a) accuracy, (b) robustness to image differences from contrast, slice thickness, and reconstruction kernels, and (c) impact of pretext tasks for pretraining. Feature reuse was evaluated using centered kernel alignment. Wild-pretrained Swin models resulted in higher feature reuse at earlier level layers and increased feature differentiation close to output. Wild-pretrained Swin outperformed self-pretrained models for analyzed imaging acquisitions. Neither ViT nor CNN showed a clear benefit of wild-pretraining compared to self-pretraining. Masked image prediction pretext task that forces networks to learn the local structure resulted in higher accuracy compared to contrastive task that models global image information. Wild-pretrained Swin networks were more robust to analyzed CT imaging differences for lung tumor segmentation than self-pretrained methods. ViT and CNN models did not show a clear benefit for wild-pretraining overself-pretraining.

Advancing cancer diagnosis and prognostication through deep learning mastery in breast, colon, and lung histopathology with ResoMergeNet

Cancer, a global health threat, demands effective diagnostic solutions to combat its impact on public health, particularly for breast, colon, and lung cancers. Early and accurate diagnosis is essential for successful treatment, prompting the rise of Computer-Aided Diagnosis Systems as reliable and cost-effective tools. Histopathology, renowned for its precision in cancer imaging, has become pivotal in the diagnostic landscape of breast, colon, and lung cancers. However, while deep learning models have been widely explored in this domain, they often face challenges in generalizing to diverse clinical settings and in efficiently capturing both local and global feature representations, particularly for multi-class tasks. This underscores the need for models that can reduce biases, improve diagnostic accuracy, and minimize error susceptibility in cancer classification tasks. To this end, we introduce ResoMergeNet (RMN), an advanced deep-learning model designed for both multi-class and binary cancer classification using histopathological images of breast, colon, and lung. ResoMergeNet integrates the Resboost mechanism which enhances feature representation, and the ConvmergeNet mechanism which optimizes feature extraction, leading to improved diagnostic accuracy. Comparative evaluations against state-of-the-art models show ResoMergeNet’s superior performance. Validated on the LC-25000 and BreakHis (400× and 40× magnifications) datasets, ResoMergeNet demonstrates outstanding performance, achieving perfect scores of 100 % in accuracy, sensitivity, precision, and F1 score for binary classification. For multi-class classification with five classes from the LC25000 dataset, it maintains an impressive 99.96 % across all performance metrics. When applied to the BreakHis dataset, ResoMergeNet achieved 99.87 % accuracy, 99.75 % sensitivity, 99.78 % precision, and 99.77 % F1 score at 400× magnification. At 40× magnification, it still delivered robust results with 98.85 % accuracy, sensitivity, precision, and F1 score. These results emphasize the efficacy of ResoMergeNet, marking a substantial advancement in diagnostic and prognostic systems for breast, colon, and lung cancers. ResoMergeNet’s superior diagnostic accuracy can significantly reduce diagnostic errors, minimize human biases, and expedite clinical workflows, making it a valuable tool for enhancing cancer diagnosis and treatment outcomes.

Development of an oral cancer detection system through deep learning

ObjectiveWe aimed to develop an AI-based model that uses a portable electronic oral endoscope to capture intraoral images of patients for the detection of oral cancer.Subjects and methodsFrom September 2019 to October 2023, 205 high-quality annotated images of oral cancer were collected using a portable oral electronic endoscope at the Chinese PLA General Hospital for this study. The U-Net and ResNet-34 deep learning models were employed for oral cancer detection. The performance of these models was evaluated using several metrics: Dice coefficient, Intersection over Union (IoU), Loss, Precision, Recall, and F1 Score.ResultsDuring the algorithm model training phase, the Dice values were approximately 0.8, the Loss values were close to 0, and the IoU values were around 0.7. In the validation phase, the highest Dice values ranged between 0.4 and 0.5, while the Loss values increased, and the training loss began to decrease gradually. In the test phase, the model achieved a maximum Precision of 0.96 with a confidence threshold of 0.990. Additionally, with a confidence threshold of 0.010, the highest F1 score reached was 0.58.ConclusionThis study provides an initial demonstration of the potential of deep learning models in identifying oral cancer.

Hu similarity coefficient: a clinically oriented metric to evaluate contour accuracy in radiation therapy

To propose a clinically oriented quantitative metric, Hu similarity coefficient (HSC), to evaluate contour quality, gauge the performance of auto contouring methods, and aid effective allocation of clinical resources. The HSC is defined as the ratio of the number of boundary points of the initial contour that doesn’t require modifications over the number of boundary points of the final adjusted contour. To demonstrate the clinical utility of the HSC in contour evaluation, we used publicly available pelvic CT data from the Cancer Imaging Archive. The bladder was selected as the organ of interest. It was contoured by a certified medical dosimetrist and reviewed by a certified medical physicist. This contour served as the ground truth contour. From this contour, we simulated two contour sets. The first set had the same Dice similarity coefficient (DSC) but different HSCs, whereas the second set kept a constant HSC while exhibiting different DSCs. Four individuals were asked to adjust the simulated contours until they met clinical standards. The corresponding contour modification times were recorded and normalized by individual’s manual contouring times from scratch. The normalized contour modification time was correlated to the HSC and DSC to evaluate their suitability as quantitative metrics assessing contour quality. The HSC maintained a strong correlation with the normalized contour modification time when both sets of simulated contours were included in analysis. The correlation between the DSC and normalized contour modification time, however, was weak. Compared to the DSC, the HSC is more suitable for evaluating contour quality. We demonstrated that the HSC correlated well with the average normalized contour modification time. Clinically, contour modification time is the most relevant factor in allocating clinical resources. Therefore, the HSC is better suited than the DSC to assess contour quality from a clinical perspective.

Molecular Engineering of Ultrabright Biomimetic NanoGhost for Site-Selective Tumor Imaging and Biodistribution.

Optically active ultrabright imaging agents are shown to delineate tumor location with deep tissue visualization in pre noclinical tumor models. NanoGhosts (NGs) particles are reconstructed from the cell membrane and integrated with organic fluorophores to attain ultra-brightness for solid tumor imaging. Moreover, the integration of amphiphilic and lipophilic molecules reveals structural characteristics of NGs (≈70nm), which also alter their brightness. Upon intravenous administration (10 mgkg-1 single dose), these ultrabright NGs (778 MESF) enable the high-resolution of tumor site and real-time tracking of vital organs with high-contrast fluorescence signals. Engineered biomimetic NGs demonstrates better resolution and tissue penetration as compared to the clinically approved indocyanine green (ICG). High precision in tumor detection (0.5h) and strong tumor retention (24h which is further up to 30th day) without affecting healthy tissues ensure the future scope of NGs in early-stage cancer imaging. These findings suggest that these NGs mimic the biological characteristics of native cells, enabling them to evade immune clearance and target the solid tumor naturally.

PSMA Accumulation in Lower Extremity Lymphedema and Hydrocele.

The development of radiolabeled small-molecule prostate-specific membrane antigen (PSMA) inhibitors has advanced molecular imaging in prostate cancer. The use of 177Lu-PSMA is especially beneficial because its therapeutic β-emission combined with γ-radiation enables precise treatment and comprehensive imaging in patients with prostate cancer. We present a case of a prostate cancer patient who developed lower extremity lymphedema and hydrocele secondary to prior radiotherapy. Posttherapy whole-body imaging following 177Lu-PSMA treatment revealed dermal backflow with a notable accumulation of soft tissue in the lower extremities and hydrocele.

An Efficient Model for Lung Cancer Detection through the Integration of Genetic Algorithm and Machine Learning

Prompt lung cancer detection is essential for patient health. Deep Learning (DL) models have been intensively used for lung cancer screening, as they provide high accuracy in diagnoses. However, DL models require significant computational power, which may not be accessible in all settings. Conventional Machine Learning (ML) models may not produce high prediction accuracy, especially with large data. This study uses a Genetic Algorithm (GA) approach to select optimal features from lung cancer images and reduce their dimensionality. This allows conventional ML models to achieve a high prediction accuracy when classifying medical images while using lower computational power compared with DL models. The proposed model integrates GA along with ML for lung cancer detection. The experimental results show that using GA with a feed-forward neural network classifier achieved high performance, reaching 99.70% classification accuracy.

Enhanced Diagnosis of Lung Cancer through an Ensemble Learning Model leveraging an Adaptive Optimization Algorithm

Early and accurate diagnosis of lung cancer is crucial to improving patient outcomes and survival rates. Machine and deep learning models have emerged as promising tools to improve the accuracy and efficiency of disease diagnosis. However, achieving optimal diagnostic performance remains a challenging task in medical research. This study integrates ensemble learning techniques with an adaptive optimization algorithm to enhance the accuracy of lung cancer diagnosis. By combining the predictive potential of multiple base classifiers, the ensemble-learning model improves overall performance and mitigates the weaknesses of individual classifiers. Additionally, the adaptive optimization algorithm dynamically adjusts the model parameters to optimize the classification performance. The effectiveness of the approach was evaluated using a comprehensive dataset that includes lung cancer images. Rigorous evaluation and comparison with state-of-the-art models showed that the proposed method achieved superior diagnostic performance, reaching an overall accuracy of 99%.

Deep learning-assisted colonoscopy images for prediction of mismatch repair deficiency in colorectal cancer.

Deficient mismatch repair or microsatellite instability is a major predictive biomarker for the efficacy of immune checkpoint inhibitors of colorectal cancer. However, routine testing has not been uniformly implemented due to cost and resource constraints. We developed and validated a deep learning-based classifiers to detect mismatch repair-deficient status from routine colonoscopy images. We obtained the colonoscopy images from the imaging database at Endoscopic Center of the Sixth Affiliated Hospital, Sun Yat-sen University. Colonoscopy images from a prospective trial (Neoadjuvant PD-1 blockade by toripalimab with or without celecoxib in mismatch repair-deficient or microsatellite instability-high locally advanced colorectal cancer) were used to test the model. A total of 5226 eligible images from 892 tumors from the consecutive patients were utilized to develop and validate the deep learning model. 2105 colorectal cancer images from 306 tumors were randomly selected to form model development dataset with a class-balanced approach. 3121 images of 488 proficient mismatch repair tumors and 98 deficient mismatch repair tumors were used to form the independent dataset. The model achieved an AUROC of 0.948 (95% CI 0.919-0.977) on the test dataset. On the independent validation dataset, the AUROC was 0.807 (0.760-0.854), and the NPV in was 94.2% (95% CI 0.918-0.967). On the prospective trial dataset, the model identified 29 tumors among the 33 deficient mismatch repair tumors (87.88%). The model achieved a high NPV in detecting deficient mismatch repair colorectal cancers. This model might serve as an automatic screening tool.

Computed Tomography Scan and Clinical-based Complete Response Prediction in Locally Advanced Rectal Cancer after Neoadjuvant Chemoradiotherapy: A Machine Learning Approach

Abstract Background: Treatment of locally advanced rectal cancer (LARC) involves neoadjuvant chemoradiotherapy (nCRT), followed by total mesorectal excision. Examining the response to treatment is one of the most important factors in the follow-up of patients; therefore, in this study, radiomics patterns derived from pretreatment computed tomography images in rectal cancer and its relationship with treatment response measurement criteria have been investigated. Methods: Fifty patients with rectal adenocarcinoma who were candidates for nCRT and surgery were included. The information obtained from the tumor surgical pathology report, including pathological T and N, the degree of tumor differentiation, lymphovascular invasion, and perineural invasion along with radiomics characteristics to each patient, was assessed. Modeling with disturbed forest model was used for radiomics data. For other variables, Shapiro–Wilk, Chi-Square, and Pearson Chi-square tests were used. Results: The participants of this study were 50 patients (23 males [46%] and 27 females [54%]). There was no significant difference in the rate of response to neoadjuvant treatment in between age and gender groups. According to the modeling based on combined clinical and radiomics data together, area under the curves for the nonresponders and complete respond group (responder group) was 0.97 and 0.99, respectively. Conclusion: Random forests modeling based on combined radiomics and clinical characteristics of the pretreatment tumor images has the ability to predict the response or non-response to neoadjuvant treatment in LARC to an acceptable extent.