Independent Test Cohort Research Articles

Quantifying impairment and disease severity using AI models trained on healthy subjects

Automatic assessment of impairment and disease severity is a key challenge in data-driven medicine. We propose a framework to address this challenge, which leverages AI models trained exclusively on healthy individuals. The COnfidence-Based chaRacterization of Anomalies (COBRA) score exploits the decrease in confidence of these models when presented with impaired or diseased patients to quantify their deviation from the healthy population. We applied the COBRA score to address a key limitation of current clinical evaluation of upper-body impairment in stroke patients. The gold-standard Fugl-Meyer Assessment (FMA) requires in-person administration by a trained assessor for 30-45 minutes, which restricts monitoring frequency and precludes physicians from adapting rehabilitation protocols to the progress of each patient. The COBRA score, computed automatically in under one minute, is shown to be strongly correlated with the FMA on an independent test cohort for two different data modalities: wearable sensors (ρ = 0.814, 95% CI [0.700,0.888]) and video (ρ = 0.736, 95% C.I [0.584, 0.838]). To demonstrate the generalizability of the approach to other conditions, the COBRA score was also applied to quantify severity of knee osteoarthritis from magnetic-resonance imaging scans, again achieving significant correlation with an independent clinical assessment (ρ = 0.644, 95% C.I [0.585,0.696]).

MRI radiomics enhances radiologists’ ability for characterizing intestinal fibrosis in patients with Crohn’s disease

ObjectivesWe aimed to develop MRI-based radiomic models (RMs) to improve the diagnostic accuracy of radiologists in characterizing intestinal fibrosis in patients with Crohn’s disease (CD).MethodsThis retrospective study included patients with refractory CD who underwent MR before surgery from November 2013 to September 2021. Resected bowel segments were histologically classified as none-mild or moderate-severe fibrosis. RMs based on different MR sequence combinations (RM1: T2WI and enhanced-T1WI; RM2: T2WI, enhanced-T1WI, diffusion-weighted imaging [DWI], and apparent diffusion coefficient [ADC]); RM3: T2WI, enhanced-T1WI, DWI, ADC, and magnetization transfer MRI [MTI]), were developed and validated in an independent test cohort. The RMs’ diagnostic performance was compared to that of visual interpretation using identical sequences and a clinical model.ResultsThe final population included 123 patients (81 men, 42 women; mean age: 30.26 ± 7.98 years; training cohort, n = 93; test cohort, n = 30). The area under the receiver operating characteristic curve (AUC) of RM1, RM2, and RM3 was 0.86 (p = 0.001), 0.88 (p = 0.001), and 0.93 (p = 0.02), respectively. The decision curve analysis confirmed a progressive improvement in the diagnostic performance of three RMs with the addition of more specific sequences. All RMs performance surpassed the visual interpretation based on the same MR sequences (visual model 1, AUC = 0.65, p = 0.56; visual model 2, AUC = 0.63, p = 0.04; visual model 3, AUC = 0.77, p = 0.002), as well as the clinical model composed of C-reactive protein and erythrocyte sedimentation rate (AUC = 0.60, p = 0.13).ConclusionsThe RMs, utilizing various combinations of conventional, DWI and MTI sequences, significantly enhance radiologists’ ability to accurately characterize intestinal fibrosis in patients with CD.Critical relevance statementThe utilization of MRI-based RMs significantly enhances the diagnostic accuracy of radiologists in characterizing intestinal fibrosis.Key PointsMRI-based RMs can characterize CD intestinal fibrosis using conventional, diffusion, and MTI sequences.The RMs achieved AUCs of 0.86–0.93 for assessing fibrosis grade.MRI-radiomics outperformed visual interpretation for grading CD intestinal fibrosis.Graphical

Machine learning prediction of prognosis in lung cancer with inter-institutional generalizability: A multicenter cohort study (WJOG15121L: REAL-WIND).

8571 Background: Recent advances in machine learning have improved the accuracy of prognosis prediction of cancer. The applicability of algorithm creation methods to data from various institutions that were not utilized for algorithm training should be established to create an ML algorithm suitable for clinical practice for patients at various facilities. However, no large-scale studies have validated the inter-institutional generalizability of this algorithm across different institutions. Methods: We conducted a multicenter, retrospective, hospital-based cohort study of consecutive patients diagnosed with stage IV lung cancer between January 2016 and December 2020. The study population was randomly categorized into the training and independent test cohorts with a 2:1 ratio. The primary metric to assess algorithm performance was the area under the receiver operating characteristic curve in the independent test cohorts. To assess the inter-institutional generalizability of the algorithm, we investigated its ability to predict patient outcomes in the remaining facility after being trained using data from 15 other facilities. Results: We enrolled 6751 patients with stage IV lung cancer from 16 institutions in Japan. The median age was 70 years, and 4,421 (65%) had a PS score of 0 or 1. As a driver oncogene, 1,515 (22%) patients had mutated EGFR. The algorithm performance metrics in the test cohort showed that the area under the receiver operating characteristic curve values ranging from 0.83 to 0.90 across different predicted survival days (180 to 1080). The performance test of 16 algorithms for investigating inter-institutional generalizability showed median areas under the receiver operating characteristic curve of 0·87 (range, 0·84–0·92), 0·84 (range, 0·78–0·88), 0·84 (range, 0·76–0·89), and 0·84 (range, 0·75–0·90) at 180, 360, 720, and 1080 days, respectively. Specifically, at 180 predicted survival days, the AUC for the performance of algorithms classified based on facilities exceeded 0.8 for all algorithms (16/16). Similarly, the AUC exceeded 0.8 for 15/16, 12/16, and 12/16 facilities at 360, 720, and 1080 days, respectively, indicating a high degree of inter-institutional generalizability of the algorithm creation method. Conclusions: This large-scale study showed that our ML algorithm can accurately predict prognosis in patients with stage IV lung cancer. Furthermore, the algorithm creation method demonstrated a high degree of inter-institutional generalizability. Therefore, our ML algorithm can be generalized to various hospitals and implemented in clinical practice to benefit patients across diverse healthcare facilities.

The use of micronuclei DNA from erythrocytes for the detection of early-stage gastrointestinal cancers.

4085 Background: Gastrointestinal (GI) cancers account for over a third of cancer deaths. However, their early detection remains a challenge. Micronuclei (MN) are extranuclear bodies containing damaged chromosome segments indicative of genomic instability. Increased MN frequency in hematopoietic cells is observed in patients with solid tumors. We have developed a method (WO2021/228246 A1) for purifying and characterizing micronuclei DNA (MN-DNA) in erythrocytes from peripheral blood. By comparing MN-DNA from healthy donors (HDs) and GI cancer patients, we identified significant changes in read densities at specific genomic locations in patients, which were termed as tumor-associated MN-DNA (taMN-DNA) signatures. Here, we evaluated the potential of these signatures for early-stage GI cancer detection, including colorectal (CC), gastric (GC) and esophageal (EC) cancers. Methods: Peripheral blood (1-2 ml) was collected from healthy donors (HD) and cancer patients MN-DNA isolation and purification from erythrocytes. Participants were randomly divided into training, validation and independent test cohorts in a 7:2:1 ratio, maintaining similar cancer types, CC/GC/EC/HD ratio, gender and age distribution. Distinct MN-DNA features between cancer patients and HDs were identified using sequencing data. Logistic regression algorithms were applied using these features for cancer detection. Results: The study enrolled 1753 participants, including 987 HDs, 420 CC, 282 GC and 64 EC cases. Of these, over half were diagnosed with early-stage diseases; TNM stage 0-I accounted for 40.57%, stage II for 23.32%, stage III for 29.25% and stage IV for 6.87%. The cancer detection model with MN-DNA features yielded an overall accuracy of 87.43%, with a sensitivity of 83.66% and specificity of 90.36%. For individual caner types, sensitivity was 85.37% for CC, 79.31% for GC and 92.31% for EC. For early-stage (stage 0-II) CC, GC and EC, the model demonstrated sensitivities of 95.12%, 86.05% and 100.00%, respectively, at 90.36% specificity. Conclusions: Unlike cell-free DNA, MN-DNA are chromosomal fragments in the cytoplasm. The abundance of erythrocytes in peripheral blood provides an easily accessible source for MN-DNA enrichment. This pilot study illustrates the potential of MN-DNA as a tool for early GI cancer detection, contributing to large-scale efforts towards developing an effective GI cancer screening test.

A multi-center study for colorectal cancer early detection among patients with high-risk disease using a cell-free fragmentomics assay.

3613 Background: Colorectal cancer (CRC) is the third most prevalent and the second deadliest cancer globally, accounting for an estimated 1,926,000 new cases and 903,800 deaths in the year 2022 alone. It's also established that chronic inflammatory conditions, such as Crohn's disease and ulcerative colitis, elevate the risk of developing CRC. Therefore, a timely CRC diagnosis for patients with colorectal disease can significantly improve their prognosis and therefore reduce cancer-related mortalities. Methods: In this study, we enrolled a total of 167 patients diagnosed with CRC (stages I: 43, II: 55, III: 56, IV: 13) and 217 patients with benign colorectal disease (BCD), from five participating hospitals. All participants underwent colonoscopy procedures, and their CRC or BCD status was pathologically confirmed. The development of the cancer detection model was based on a training cohort of 98 CRC patients and 131 BCD patients from three hospitals, and the evaluation of the model's performance was carried out on an independent test cohort comprising 69 CRC patients and 96 BCD patients from the remaining two hospitals. The predictive model leveraged cell-free fragmentomics profiling, employing low-pass whole-genome sequencing (WGS) on pre-operative plasma samples to identify CRC. Results: The early detection model exhibited remarkable efficacy in differentiating CRC patients from those with BCD, achieving an Area Under the Curve (AUC) of 0.930 in the training cohort and a nearly equivalent AUC of 0.926 in the independent test cohort. By applying the same cutoff determined in the training cohort, our model demonstrated a high sensitivity (91.8% in the training cohort and 91.3% in the test cohort) and a satisfactory specificity (78.6% in the training cohort and 82.3% in the test cohort). Notably, the model showed consistent and robust performance across various CRC characteristics, including tumor location, histology subtypes, grades, and mismatch repair deficiency status in both cohorts. Moreover, our model outperformed traditional CRC screening methods such as the fecal immunochemical test, carcinoembryonic antigen (CEA), and CA19-9, which achieved AUCs of 0.742, 0.720, and 0.543, respectively. Further validation was conducted using an additional cohort of 31 patients with advanced colorectal adenoma (advCRA), where the model achieved an AUC of 0.846 and a 66.7% sensitivity for distinguishing advCRA from BCD. Conclusions: The study introduces a non-invasive, cell-free fragmentomics assay that incorporates low-pass WGS and machine learning algorithms, demonstrating significant clinical potential in accurately distinguishing CRC and BCD patients. Our innovative approach offers a promising alternative to traditional screening methods that can improve early CRC detection, enhance CRC patient outcomes, and reduce mortality rates.

A multi-modal artificial intelligence model trained on H&E tissue images and clinical patient data for prediction of BRAF mutations in colorectal cancer.

e15161 Background: Analyzing genetic mutations is essential for personalized therapy decisions in oncology and beyond. In colorectal cancer (CRC), mutations in BRAF ( BRAFmut) render patients eligible for targeted treatment with BRAF inhibitors and immunotherapy. BRAFmut is routinely analyzed via genomic sequencing, which, however, is costly and does not provide diagnostic results immediately. Reliable detection of BRAFmut status from histological images using artificial intelligence (AI) could provide a more accessible and faster diagnostic alternative. Previous approaches to image-based BRAFmut analysis, however, failed to reach a critical level of accuracy on independent test cohorts, with reported area under the receiver operating characteristic curve (AUROC) values typically below 0.8. Methods: We hypothesized that combining tissue image analysis with clinical patient data could improve diagnostic accuracy for BRAFmut. Following our hypothesis, we developed a multi-modal AI model on n = 426 CRC cases from the TCGA database using BRAFmut data, H&E-stained tissue images, and information on patient parameters that are readily available in clinical practice and that are known to be associated with BRAFmut frequency. The model was based on a vision transformer architecture incorporating pathology-specific information using a pretrained embedding model for encoding tissue image features. Model performance was validated on an independent hold-out TCGA cohort of n = 107 samples and an additional external cohort of n = 104 CRC samples from the CPTAC database. Results: Our multi-modal AI model reached an AUROC of 0.85 ± 0.02 on the TCGA hold-out test set. Notably, on the second, external CPTAC dataset the predictive accuracy level was similar (AUROC of 0.83 ± 0.02) indicating the robustness of our approach to generalize to different datasets. Additional analyses showed that the significant performance boost over previous approaches was gained by the multi-modal analysis setup, as predicting BRAFmut based on either tissue, or clinical data alone led to lower accuracy values. Explainability analysis of AI results confirmed that relevant biological features on the histological slides, mainly tumor areas, were considered by the model for decision making. Conclusions: The combination of H&E images with clinically readily available patient parameters enabled us to develop a model predicting BRAFmut in CRC that surpasses previous approaches in diagnostic accuracy on independent evaluation cohorts. This indicates that combining histological tissue information with clinical parameters can increase the ability of AI models to predict genetic variants from tissue and more generally highlights the potential of multi-modal deep learning approaches to enhance predictive AI models in digital pathology.

MRI-based intratumoral and peritumoral radiomics for preoperative prediction of glioma grade: a multicenter study.

Accurate preoperative prediction of glioma is crucial for developing individualized treatment decisions and assessing prognosis. In this study, we aimed to establish and evaluate the value of integrated models by incorporating the intratumoral and peritumoral features from conventional MRI and clinical characteristics in the prediction of glioma grade. A total of 213 glioma patients from two centers were included in the retrospective analysis, among which, 132 patients were classified as the training cohort and internal validation set, and the remaining 81 patients were zoned as the independent external testing cohort. A total of 7728 features were extracted from MRI sequences and various volumes of interest (VOIs). After feature selection, 30 radiomic models depended on five sets of machine learning classifiers, different MRI sequences, and four different combinations of predictive feature sources, including features from the intratumoral region only, features from the peritumoral edema region only, features from the fusion area including intratumoral and peritumoral edema region (VOI-fusion), and features from the intratumoral region with the addition of features from peritumoral edema region (feature-fusion), were established to select the optimal model. A nomogram based on the clinical parameter and optimal radiomic model was constructed for predicting glioma grade in clinical practice. The intratumoral radiomic models based on contrast-enhanced T1-weighted and T2-flair sequences outperformed those based on a single MRI sequence. Moreover, the internal validation and independent external test underscored that the XGBoost machine learning classifier, incorporating features extracted from VOI-fusion, showed superior predictive efficiency in differentiating between low-grade gliomas (LGG) and high-grade gliomas (HGG), with an AUC of 0.805 in the external test. The radiomic models of VOI-fusion yielded higher prediction efficiency than those of feature-fusion. Additionally, the developed nomogram presented an optimal predictive efficacy with an AUC of 0.825 in the testing cohort. This study systematically investigated the effect of intratumoral and peritumoral radiomics to predict glioma grading with conventional MRI. The optimal model was the XGBoost classifier coupled radiomic model based on VOI-fusion. The radiomic models that depended on VOI-fusion outperformed those that depended on feature-fusion, suggesting that peritumoral features should be rationally utilized in radiomic studies.

Deep learning-based whole-body PSMA PET/CT attenuation correction utilizing Pix-2-Pix GAN.

Sequential PET/CT studies oncology patients can undergo during their treatment follow-up course is limited by radiation dosage. We propose an artificial intelligence (AI) tool to produce attenuation-corrected PET (AC-PET) images from non-attenuation-corrected PET (NAC-PET) images to reduce need for low-dose CT scans. A deep learning algorithm based on 2D Pix-2-Pix generative adversarial network (GAN) architecture was developed from paired AC-PET and NAC-PET images. 18F-DCFPyL PSMA PET-CT studies from 302 prostate cancer patients, split into training, validation, and testing cohorts (n = 183, 60, 59, respectively). Models were trained with two normalization strategies: Standard Uptake Value (SUV)-based and SUV-Nyul-based. Scan-level performance was evaluated by normalized mean square error (NMSE), mean absolute error (MAE), structural similarity index (SSIM), and peak signal-to-noise ratio (PSNR). Lesion-level analysis was performed in regions-of-interest prospectively from nuclear medicine physicians. SUV metrics were evaluated using intraclass correlation coefficient (ICC), repeatability coefficient (RC), and linear mixed-effects modeling. Median NMSE, MAE, SSIM, and PSNR were 13.26%, 3.59%, 0.891, and 26.82, respectively, in the independent test cohort. ICC for SUVmax and SUVmean were 0.88 and 0.89, which indicated a high correlation between original and AI-generated quantitative imaging markers. Lesion location, density (Hounsfield units), and lesion uptake were all shown to impact relative error in generated SUV metrics (all p < 0.05). The Pix-2-Pix GAN model for generating AC-PET demonstrates SUV metrics that highly correlate with original images. AI-generated PET images show clinical potential for reducing the need for CT scans for attenuation correction while preserving quantitative markers and image quality.

Machine learning model for prediction of permanent stoma after anterior resection of rectal cancer: A multicenter study

BackgroundThe conversion from a temporary to a permanent stoma (PS) following rectal cancer surgery significantly impacts the quality of life of patients. However, there is currently a lack of practical preoperative tools to predict PS formation. The purpose of this study is to establish a preoperative predictive model for PS using machine learning algorithms to guide clinical practice. MethodsIn this retrospective study, we analyzed clinical data from a total of 655 patients who underwent anterior resection for rectal cancer, with 552 patients from one medical center and 103 from another. Through machine learning algorithms, five predictive models were developed, and each was thoroughly evaluated for predictive performance. The model with superior predictive accuracy underwent additional validation using both an independent testing cohort and the external validation cohort. The Shapley Additive exPlanations (SHAP) approach was employed to elucidate the predictive factors influencing the model, providing an in-depth visual analysis of its decision-making process. ResultsEight variables were selected for the construction of the model. The support vector machine (SVM) model exhibited superior predictive performance in the training set, evidenced by an AUC of 0.854 (95 % CI:0.803–0.904). This performance was corroborated in both the testing set and external validation set, where the model demonstrated an AUC of 0.851 (95%CI:0.748–0.954) and 0.815 (95%CI:0.710–0.919), respectively, indicating its efficacy in identifying the PS. ConclusionsThe model(https://yangsu2023.shinyapps.io/psrisk/) indicated robust predictive performance in identifying PS after anterior resection for rectal cancer, potentially guiding surgeons in the preoperative stratification of patients, thus informing individualized treatment plans and improving patient outcomes.

Predicting lung function decline in cystic fibrosis: the impact of initiating ivacaftor therapy

BackgroundModulator therapies that seek to correct the underlying defect in cystic fibrosis (CF) have revolutionized the clinical landscape. Given the heterogeneous nature of lung disease progression in the post-modulator era, there is a need to develop prediction models that are robust to modulator uptake.MethodsWe conducted a retrospective longitudinal cohort study of the CF Foundation Patient Registry (N = 867 patients carrying the G551D mutation who were treated with ivacaftor from 2003 to 2018). The primary outcome was lung function (percent predicted forced expiratory volume in 1 s or FEV1pp). To characterize the association between ivacaftor initiation and lung function, we developed a dynamic prediction model through covariate selection of demographic and clinical characteristics. The ability of the selected model to predict a decline in lung function, clinically known as an FEV1-indicated exacerbation signal (FIES), was evaluated both at the population level and individual level.ResultsBased on the final model, the estimated improvement in FEV1pp after ivacaftor initiation was 4.89% predicted (95% confidence interval [CI]: 3.90 to 5.89). The rate of decline was reduced with ivacaftor initiation by 0.14% predicted/year (95% CI: 0.01 to 0.27). More frequent outpatient visits prior to study entry and being male corresponded to a higher overall FEV1pp. Pancreatic insufficiency, older age at study entry, a history of more frequent pulmonary exacerbations, lung infections, CF-related diabetes, and use of Medicaid insurance corresponded to lower FEV1pp. The model had excellent predictive accuracy for FIES events with an area under the receiver operating characteristic curve of 0.83 (95% CI: 0.83 to 0.84) for the independent testing cohort and 0.90 (95% CI: 0.89 to 0.90) for 6-month forecasting with the masked cohort. The root-mean-square errors of the FEV1pp predictions for these cohorts were 7.31% and 6.78% predicted, respectively, with standard deviations of 0.29 and 0.20. The predictive accuracy was robust across different covariate specifications.ConclusionsThe methods and applications of dynamic prediction models developed using data prior to modulator uptake have the potential to inform post-modulator projections of lung function and enhance clinical surveillance in the new era of CF care.

Radiomics and machine learning for renal tumor subtype assessment using multiphase computed tomography in a multicenter setting

ObjectivesTo distinguish histological subtypes of renal tumors using radiomic features and machine learning (ML) based on multiphase computed tomography (CT).Material and methodsPatients who underwent surgical treatment for renal tumors at two tertiary centers from 2012 to 2022 were included retrospectively. Preoperative arterial (corticomedullary) and venous (nephrogenic) phase CT scans from these centers, as well as from external imaging facilities, were manually segmented, and standardized radiomic features were extracted. Following preprocessing and addressing the class imbalance, a ML algorithm based on extreme gradient boosting trees (XGB) was employed to predict renal tumor subtypes using 10-fold cross-validation. The evaluation was conducted using the multiclass area under the receiver operating characteristic curve (AUC). Algorithms were trained on data from one center and independently tested on data from the other center.ResultsThe training cohort comprised n = 297 patients (64.3% clear cell renal cell cancer [RCC], 13.5% papillary renal cell carcinoma (pRCC), 7.4% chromophobe RCC, 9.4% oncocytomas, and 5.4% angiomyolipomas (AML)), and the testing cohort n = 121 patients (56.2%/16.5%/3.3%/21.5%/2.5%).The XGB algorithm demonstrated a diagnostic performance of AUC = 0.81/0.64/0.8 for venous/arterial/combined contrast phase CT in the training cohort, and AUC = 0.75/0.67/0.75 in the independent testing cohort. In pairwise comparisons, the lowest diagnostic accuracy was evident for the identification of oncocytomas (AUC = 0.57–0.69), and the highest for the identification of AMLs (AUC = 0.9–0.94)ConclusionRadiomic feature analyses can distinguish renal tumor subtypes on routinely acquired CTs, with oncocytomas being the hardest subtype to identify.Clinical relevance statementRadiomic feature analyses yield robust results for renal tumor assessment on routine CTs. Although radiologists routinely rely on arterial phase CT for renal tumor assessment and operative planning, radiomic features derived from arterial phase did not improve the accuracy of renal tumor subtype identification in our cohort.

Plasma proteomic characterization of colorectal cancer patients with FOLFOX chemotherapy by integrated proteomics technology

BackgroundColorectal Cancer (CRC) is a prevalent form of cancer, and the effectiveness of the main postoperative chemotherapy treatment, FOLFOX, varies among patients. In this study, we aimed to identify potential biomarkers for predicting the prognosis of CRC patients treated with FOLFOX through plasma proteomic characterization.MethodsUsing a fully integrated sample preparation technology SISPROT-based proteomics workflow, we achieved deep proteome coverage and trained a machine learning model from a discovery cohort of 90 CRC patients to differentiate FOLFOX-sensitive and FOLFOX-resistant patients. The model was then validated by targeted proteomics on an independent test cohort of 26 patients.ResultsWe achieved deep proteome coverage of 831 protein groups in total and 536 protein groups in average for non-depleted plasma from CRC patients by using a Orbitrap Exploris 240 with moderate sensitivity. Our results revealed distinct molecular changes in FOLFOX-sensitive and FOLFOX-resistant patients. We confidently identified known prognostic biomarkers for colorectal cancer, such as S100A4, LGALS1, and FABP5. The classifier based on the biomarker panel demonstrated a promised AUC value of 0.908 with 93% accuracy. Additionally, we established a protein panel to predict FOLFOX effectiveness, and several proteins within the panel were validated using targeted proteomic methods.ConclusionsOur study sheds light on the pathways affected in CRC patients treated with FOLFOX chemotherapy and identifies potential biomarkers that could be valuable for prognosis prediction. Our findings showed the potential of mass spectrometry-based proteomics and machine learning as an unbiased and systematic approach for discovering biomarkers in CRC.

Blood leukocytes as a non-invasive diagnostic tool for thyroid nodules: a prospective cohort study

BackgroundThyroid nodule (TN) patients in China are subject to overdiagnosis and overtreatment. The implementation of existing technologies such as thyroid ultrasonography has indeed contributed to the improved diagnostic accuracy of TNs. However, a significant issue persists, where many patients undergo unnecessary biopsies, and patients with malignant thyroid nodules (MTNs) are advised to undergo surgery therapy.MethodsThis study included a total of 293 patients diagnosed with TNs. Differential methylation haplotype blocks (MHBs) in blood leukocytes between MTNs and benign thyroid nodules (BTNs) were detected using reduced representation bisulfite sequencing (RRBS). Subsequently, an artificial intelligence blood leukocyte DNA methylation (BLDM) model was designed to optimize the management and treatment of patients with TNs for more effective outcomes.ResultsThe DNA methylation profiles of peripheral blood leukocytes exhibited distinctions between MTNs and BTNs. The BLDM model we developed for diagnosing TNs achieved an area under the curve (AUC) of 0.858 in the validation cohort and 0.863 in the independent test cohort. Its specificity reached 90.91% and 88.68% in the validation and independent test cohorts, respectively, outperforming the specificity of ultrasonography (43.64% in the validation cohort and 47.17% in the independent test cohort), albeit with a slightly lower sensitivity (83.33% in the validation cohort and 82.86% in the independent test cohort) compared to ultrasonography (97.62% in the validation cohort and 100.00% in the independent test cohort). The BLDM model could correctly identify 89.83% patients whose nodules were suspected malignant by ultrasonography but finally histological benign. In micronodules, the model displayed higher specificity (93.33% in the validation cohort and 92.00% in the independent test cohort) and accuracy (88.24% in the validation cohort and 87.50% in the independent test cohort) for diagnosing TNs. This performance surpassed the specificity and accuracy observed with ultrasonography. A TN diagnostic and treatment framework that prioritizes patients is provided, with fine-needle aspiration (FNA) biopsy performed only on patients with indications of MTNs in both BLDM and ultrasonography results, thus avoiding unnecessary biopsies.ConclusionsThis is the first study to demonstrate the potential of non-invasive blood leukocytes in diagnosing TNs, thereby making TN diagnosis and treatment more efficient in China.

Abstract 6184: A novel approach to Barrett's esophagus risk stratification: Whole-slide image analysis for aneuploidy detection

Abstract Background Barrett’s esophagus (BE) is the only known precursor of esophageal adenocarcinoma (EAC) and there is a need for biomarkers in BE for risk stratification for cancer progression. Aneuploidy has been suggested as a factor in the development, initiation or progression of EAC in patients with BE, and it has been found predictive for EAC progression (Hadjinicolaou, et al. 2020, Sikkema, et al. 2009). However, current assays for detecting aneuploidy require valuable tissue, hence validation studies are limited. We hypothesise that routine histology images can be used to detect ploidy status. We aim to develop a sensitive, accurate, and easy-to-use deep learning tool for this purpose. Methods We used a weakly supervised deep learning-based approach called clustering-constrained-attention multiple-instance learning (CLAM) (Lu et al., 2021) to detect aneuploidy on hematoxylin and eosin-stained whole slide images of endoscopical biopsies from BE, for which the ploidy status had been determined by flow cytometry. To benchmark the model’s performance, we also trained a traditional fully supervised algorithm, ResNet50 (He et al., 2016), with the same dataset. The models were trained on 388 slides (51 aneuploid) from the Seattle Barrett’s Esophagus Annotated Resource (BEAR) and then applied to an independent test cohort of BEAR patients, consisting of 279 slides (36 aneuploid). Results The multi-attention-branch CLAM model achieved AUC of 0.82 and 76.4% balanced accuracy for aneuploidy on the internal test subset (10% of the cohort). On the independent test dataset, the model achieved AUC of 0.85 and 79.3% balanced accuracy, while performance of the fully supervised model was AUC of 0.65 and 33.1 balanced accuracy. In a challenging subset for image-based diagnosis, including 253 samples (29 aneuploid) without dysplasia or atypical mitosis (a major feature of aneuploidy samples) the model achieved a 67.6% accuracy and an AUC of 0.83. Both the flow results (p=2.84e-7) and the model's predictions (p=2.18e-3) revealed a correlation between progression to EAC and ploidy status. Conclusion We developed a deep learning model for predicting aneuploidy in BE biopsies. The weakly-supervised approach can perform much better compared to the traditional fully-supervised model. Aneuploidy is correlated with cancer progression and stands as a promising candidate for the prediction of cancer progression in BE patients. Although atypical mitosis is the primary histological indicator of aneuploidy, our model can still make aneuploid predictions in the absence of atypical mitosis, as it doesn't rely solely on a single parameter. Our model is efficient, capable of processing hundreds of samples resource-effectively, and thus an ideal adjunct to standard histologic evaluation. This classifier may facilitate molecular-based improvements in identifying individuals at high risk of progression. Citation Format: Caner Ercan, Xiaoxi Pan, Thomas G. Paulson, Matthew D. Stachler, Carlo C. Maley, Yinyin Yuan. A novel approach to Barrett's esophagus risk stratification: Whole-slide image analysis for aneuploidy detection [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6184.

An ultrasound-based nomogram for predicting axillary node pathologic complete response after neoadjuvant chemotherapy in breast cancer: Modeling and external validation.

The staging and treatment of axillary nodes in breast cancer have become a focus of research. For breast cancer patients with fine-needle aspiration-or core needle biopsy-confirmed positive nodes, axillary lymph node dissection (ALND) after neoadjuvant chemotherapy (NAC) is still a standard treatment. However, some patients achieve an axillary pathologic complete response (pCR) after NAC. In this study, the authors sought to construct a model to predict axillary pCR in patients with positive axillary lymph nodes (cN+) breast cancer. Data from patients with pathologically proven cN+ breast cancer treated with NAC followed by ALND between January 2010 and April 2019 at the Peking University Cancer Hospital were reviewed. Axillary lymph node status was assessed using ultrasonography before and after NAC. The patient cohort was assigned to the construction and internal validation cohorts according to admission time. A nomogram was constructed based on the significant factors associated with axillary pCR. The predictive performance of the model was externally validated using data from Peking University First Hospital. This study included 953 and 267 patients from Peking University Cancer Hospital and Peking University First Hospital, respectively. In the construction cohort, 39.7% (238 of 600) of patients achieved axillary pCR after NAC. The result of multivariate logistic regression analysis showed that tumor grade, clinical nodal response, NAC regimen, tumor pCR, lymphovascular invasion, and tumor biologic subtype were significant independent predictors of ypN0 (p < 0.05). The areas under the receiver operating characteristic curves for the construction, validation, and independent testing cohorts were 0.87 (95% confidence interval [CI], 0.84-0.90), 0.83 (95% CI, 0.79-0.87), and 0.84 (0.79-0.89), respectively. A nomogram was constructed to predict the pCR of axillary lymph nodes after NAC for breast cancer. Validation of both the internal and external cohorts achieved good predictive performance, indicating that the model has preliminary clinical application prospects.

Machine Learning Prediction of Treatment Response to Inhaled Corticosteroids in Asthma.

Although inhaled corticosteroids (ICS) are the first-line therapy for patients with persistent asthma, many patients continue to have exacerbations. We developed machine learning models to predict the ICS response in patients with asthma. The subjects included asthma patients of European ancestry (n = 1371; 448 children; 916 adults). A genome-wide association study was performed to identify the SNPs associated with ICS response. Using the SNPs identified, two machine learning models were developed to predict ICS response: (1) least absolute shrinkage and selection operator (LASSO) regression and (2) random forest. The LASSO regression model achieved an AUC of 0.71 (95% CI 0.67-0.76; sensitivity: 0.57; specificity: 0.75) in an independent test cohort, and the random forest model achieved an AUC of 0.74 (95% CI 0.70-0.78; sensitivity: 0.70; specificity: 0.68). The genes contributing to the prediction of ICS response included those associated with ICS responses in asthma (TPSAB1, FBXL16), asthma symptoms and severity (ABCA7, CNN2, PTRN3, and BSG/CD147), airway remodeling (ELANE, FSTL3), mucin production (GAL3ST), leukotriene synthesis (GPX4), allergic asthma (ZFPM1, SBNO2), and others. An accurate risk prediction of ICS response can be obtained using machine learning methods, with the potential to inform personalized treatment decisions. Further studies are needed to examine if the integration of richer phenotype data could improve risk prediction.

Radiomics-Based Prediction of Collateral Status from CT Angiography of Patients Following a Large Vessel Occlusion Stroke.

A major driver of individual variation in long-term outcomes following a large vessel occlusion (LVO) stroke is the degree of collateral arterial circulation. We aimed to develop and evaluate machine-learning models that quantify LVO collateral status using admission computed tomography angiography (CTA) radiomics. We extracted 1116 radiomic features from the anterior circulation territories from admission CTAs of 600 patients experiencing an acute LVO stroke. We trained and validated multiple machine-learning models for the prediction of collateral status based on consensus from two neuroradiologists as ground truth. Models were first trained to predict (1) good vs. intermediate or poor, or (2) good vs. intermediate or poor collateral status. Then, model predictions were combined to determine a three-tier collateral score (good, intermediate, or poor). We used the receiver operating characteristics area under the curve (AUC) to evaluate prediction accuracy. We included 499 patients in training and 101 in an independent test cohort. The best-performing models achieved an averaged cross-validation AUC of 0.80 ± 0.05 for poor vs. intermediate/good collateral and 0.69 ± 0.05 for good vs. intermediate/poor, and AUC = 0.77 (0.67-0.87) and AUC = 0.78 (0.70-0.90) in the independent test cohort, respectively. The collateral scores predicted by the radiomics model were correlated with (rho = 0.45, p = 0.002) and were independent predictors of 3-month clinical outcome (p = 0.018) in the independent test cohort. Automated tools for the assessment of collateral status from admission CTA-such as the radiomics models described here-can generate clinically relevant and reproducible collateral scores to facilitate a timely treatment triage in patients experiencing an acute LVO stroke.

X-rays radiomics-based machine learning classification of atypical cartilaginous tumour and high-grade chondrosarcoma of long bones

Atypical cartilaginous tumour (ACT) and high-grade chondrosarcoma (CS) of long bones are respectively managed with active surveillance or curettage and wide resection. Our aim was to determine diagnostic performance of X-rays radiomics-based machine learning for classification of ACT and high-grade CS of long bones. This retrospective, IRB-approved study included 150 patients with surgically treated and histology-proven lesions at two tertiary bone sarcoma centres. At centre 1, the dataset was split into training (n=71 ACT, n=24 high-grade CS) and internal test (n=19 ACT, n=6 high-grade CS) cohorts, respectively, based on the date of surgery. At centre 2, the dataset constituted the external test cohort (n=12 ACT, n=18 high-grade CS). Manual segmentation was performed on frontal view X-rays, using MRI or CT for preliminary identification of lesion margins. After image pre-processing, radiomic features were extracted. Dimensionality reduction included stability, coefficient of variation, and mutual information analyses. In the training cohort, after class balancing, a machine learning classifier (Support Vector Machine) was automatically tuned using nested 10-fold cross-validation. Then, it was tested on both the test cohorts and compared to two musculoskeletal radiologists' performance using McNemar's test. Five radiomic features (3 morphology, 2 texture) passed dimensionality reduction. After tuning on the training cohort (AUC=0.75), the classifier had 80%, 83%, 79% and 80%, 89%, 67% accuracy, sensitivity, and specificity in the internal (temporally independent) and external (geographically independent) test cohorts, respectively, with no difference compared to the radiologists (p≥0.617). X-rays radiomics-based machine learning accurately differentiates between ACT and high-grade CS of long bones. AIRC Investigator Grant.

Deep-learning model to improve histological grading and predict upstaging of atypical ductal hyperplasia / ductal carcinoma in situ on breast biopsy.

Risk stratification of atypical ductal hyperplasia (ADH) and ductal carcinoma in situ (DCIS), diagnosed using breast biopsy, has great clinical significance. Clinical trials are currently exploring the possibility of active surveillance for low-risk lesions, whereas axillary lymph node staging may be considered during surgical planning for high-risk lesions. We aimed to develop a machine-learning algorithm based on whole-slide images of breast biopsy specimens and clinical information to predict the risk of upstaging to invasive breast cancer after wide excision. Patients diagnosed with ADH/DCIS on breast biopsy were included in this study, comprising 592 (740 slides) and 141 (198 slides) patients in the development and independent testing cohorts, respectively. Histological grading of the lesions was independently evaluated by two pathologists. Clinical information, including biopsy method, lesion size, and Breast Imaging Reporting and Data System (BI-RADS) classification of ultrasound and mammograms, were collected. Deep DCIS consisted of three deep neural networks to evaluate nuclear grade, necrosis, and stromal reactivity. Deep DCIS output comprised five parameters: total patches, lesion extent, Deep Grade, Deep Necrosis, and Deep Stroma. Deep DCIS highly correlated with the pathologists' evaluations of both slide- and patient-level labels. All five parameters of Deep DCIS were significantly associated with upstaging to invasive carcinoma in subsequent wide excisional specimens. Using multivariate logistic regression, Deep DCIS predicted upstaging to invasive carcinoma with an area under the curve (AUC) of 0.81, outperforming pathologists' evaluation (AUC, 0.71 and 0.69). After including clinical and hormone receptor status information, performance further improved (AUC, 0.87). This combined model retained its predictive power in two subgroup analyses: the first subgroup included unequivocal DCIS (excluding cases of ADH and DCIS suspicious for microinvasion) (AUC, 0.83), while the second excluded cases of high-grade DCIS (AUC, 0.81). The model was validated in an independent testing cohort (AUC, 0.81). This study demonstrated that deep-learning models can refine histological evaluation of ADH and DCIS on breast biopsies, which may help guide future treatment planning.

Complemental Value of Microstructural and Macrostructural MRI in the Discrimination of Neurodegenerative Parkinson Syndromes.

Various MRI-based techniques were tested for the differentiation of neurodegenerative Parkinson syndromes (NPS); the value of these techniques in direct comparison and combination is uncertain. We thus compared the diagnostic performance of macrostructural, single compartmental, and multicompartmental MRI in the differentiation of NPS. We retrospectively included patients with NPS, including 136 Parkinson's disease (PD), 41 multiple system atrophy (MSA) and 32 progressive supranuclear palsy (PSP) and 27healthy controls (HC). Macrostructural tissue probability values (TPV) were obtained by CAT12. The microstructure was assessed using amesoscopic approach by diffusion tensor imaging (DTI), neurite orientation dispersion and density imaging (NODDI), and diffusion microstructure imaging (DMI). After an atlas-based read-out, alinear support vector machine (SVM) was trained on atraining set (n = 196) and validated in an independent test cohort (n = 40). The diagnostic performance of the SVM was compared for different inputs individually and in combination. Regarding the inputs separately, we observed the best diagnostic performance for DMI. Overall, the combination of DMI and TPV performed best and correctly classified 88% of the patients. The corresponding area under the receiver operating characteristic curve was 0.87 for HC, 0.97 for PD, 1.0 for MSA, and 0.99 for PSP. We were able to demonstrate that (1) MRI parameters that approximate the microstructure provided substantial added value over conventional macrostructural imaging, (2) multicompartmental biophysically motivated models performed better than the single compartmental DTI and (3) combining macrostructural and microstructural information classified NPS and HC with satisfactory performance, thus suggesting acomplementary value of both approaches.