Annotated Regions Of Interest Research Articles

The use of deep learning state-of-the-art architectures for oral epithelial dysplasia grading: A comparative appraisal.

Dysplasia grading systems for oral epithelial dysplasia are a source of disagreement among pathologists. Therefore, machine learning approaches are being developed to mitigate this issue. This cross-sectional study included a cohort of 82 patients with oral potentially malignant disorders and correspondent 98 hematoxylin and eosin-stained whole slide images with biopsied-proven dysplasia. All whole-slide images were manually annotated based on the binary system for oral epithelial dysplasia. The annotated regions of interest were segmented and fragmented into small patches and non-randomly sampled into training/validation and test subsets. The training/validation data were color augmented, resulting in a total of 81,786 patches for training. The held-out independent test set enrolled a total of 4,486 patches. Seven state-of-the-art convolutional neural networks were trained, validated, and tested with the same dataset. The models presented a high learning rate, yet very low generalization potential. At the model development, VGG16 performed the best, but with massive overfitting. In the test set, VGG16 presented the best accuracy, sensitivity, specificity, and area under the curve (62%, 62%, 66%, and 65%, respectively), associated with the higher loss among all Convolutional Neural Networks (CNNs) tested. EfficientB0 has comparable metrics and the lowest loss among all convolutional neural networks, being a great candidate for further studies. The models were not able to generalize enough to be applied in real-life datasets due to an overlapping of features between the two classes (i.e., high risk and low risk of malignization).

AI co-pilot: content-based image retrieval for the reading of rare diseases in chest CT

The aim of the study was to evaluate the impact of the newly developed Similar patient search (SPS) Web Service, which supports reading complex lung diseases in computed tomography (CT), on the diagnostic accuracy of residents. SPS is an image-based search engine for pre-diagnosed cases along with related clinical reference content (https://eref.thieme.de). The reference database was constructed using 13,658 annotated regions of interest (ROIs) from 621 patients, comprising 69 lung diseases. For validation, 50 CT scans were evaluated by five radiology residents without SPS, and three months later with SPS. The residents could give a maximum of three diagnoses per case. A maximum of 3 points was achieved if the correct diagnosis without any additional diagnoses was provided. The residents achieved an average score of 17.6 ± 5.0 points without SPS. By using SPS, the residents increased their score by 81.8% to 32.0 ± 9.5 points. The improvement of the score per case was highly significant (p = 0.0001). The residents required an average of 205.9 ± 350.6 s per case (21.9% increase) when SPS was used. However, in the second half of the cases, after the residents became more familiar with SPS, this increase dropped to 7%. Residents’ average score in reading complex chest CT scans improved by 81.8% when the AI-driven SPS with integrated clinical reference content was used. The increase in time per case due to the use of the SPS was minimal.

Abstract 2756: A toolbox enabling a data-driven selection of regions of interest in tumor tissue sections for imaging mass cytometry analysis

Abstract Highly multiplexed imaging technologies are increasingly applied in biomedical research to identify underlying patterns of disease. As part of an IMI-funded project, the IMMUcan (Integrated immunoprofiling of large adaptive cancer patients cohorts) consortium will generate single-cell resolved molecular profiles for thousands of cancer patients from hospitals across Europe. Patient tissues will be profiled using two imaging modalities, multispectral immunofluorescence (mIF) and Imaging Mass Cytometry (IMC), in junction with data from different omics techniques. IMC enables highly multiplexed tissue imaging, but its current throughput only enables to image a fraction of large/whole-slide tumor tissue sections. The selection of regions of interest for IMC analysis is of critical importance with respect to the cells present in an area and the antibody panel used for the analysis of such an area. To target areas of interest in tumor tissue sections, e.g. with immune cell infiltration or tertiary lymphoid structures, we have developed a data driven approach based on mIF imaging analyses from consecutive sections to select such regions of interest for analysis with IMC. A mIF panel was designed to stain T cells, B cells, macrophages, neutrophils, dendritic cells and tumor cells to have an overview of immune cell populations within the whole tissue. After quantification, mIF images and data are hosted on a webserver and regions that contain cell-types of interest are annotated with the help of a web-based tool. mIF images are used for alignment with brightfield whole slide scans of the consecutive FFPE sections which are used for IMC. After alignment, the annotated regions of interest are acquired in IMC. The mean positional offset from IF to IMC for regions of interest was roughly 120 µm and the overall cell-type numbers across both technologies were well correlated. The approach presented here has the great advantage to allow us to reproducibly target similar areas for IMC across thousands of samples based on mIF within the IMMUcan consortium and will enable single-cell analysis for whole tumor sections based on mIF and IMC to gain a deeper understanding of the underlying tumor biology. Funding: IMI2 JU grant agreement 821558, supported by EU's Horizon 2020 and EFPIA. Citation Format: Daniel Schulz, Stephanie Renaud-Tissot, Robin Liechti, Nils Eling, Michelle Daniel, Jonas Windhager, Yervand Karapetyan, Thomas Mrowiec, Sylvie Rusakiewicz, Henoch S. Hong, Marie Morfouace, Bernd Bodenmiller. A toolbox enabling a data-driven selection of regions of interest in tumor tissue sections for imaging mass cytometry analysis [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 2756.

Abstract 193: A deep learning workflow to quantify the H-Score on immunohistochemistry images

Abstract Quantification of protein expression in IHC is typically assessed with H-scores, calculated by a pathologist on a scale from 0 to 300 based on the intensity of carcinoma cells and their relative proportion.The subjectivity of the methodology is an inherent risk of inconsistency for inter- intra- assay. Here, we developed a semi-automated workflow based on deep learning to calculate the H-score on IHC slides for the CDH17 protein, a component of the gastrointestinal (GI) system. CDH17 membranous expression makes its quantification particularly challenging and prone to variability. This workflow was validated on 112 samples of GI tumors (primary as well as metastatic), reaching an H-score correlation score R2=0.92 (p-value below 10-5) and an overall median absolute error=17.5 compared to a pathologist's evaluation. In our method, tumor regions on each slide image were first annotated by a pathologist, containing clusters of epithelial carcinoma mixed with non-tumor tissue. A segmentation model was trained to outline contours of marked carcinoma clusters, later classified as a low, moderate or strong expression based on optical density distribution. Finally, all cluster results were aggregated at slide level to get a final H-score. The method was evaluated on 112 tumor sections including the following cancer histotypes: pancreatic (N=20), gastric (N=18), esophageal (N=20), cholangiocarcinoma (N=20) and colorectal (N=34). A U-Net segmentation model with a ResNet101 encoder was trained on 303 annotated regions of interest (ROIs) of size 240x240 µm2 taken from 6 images, scanned at 20x. The model was taught to segment CDH17- and CDH17+ carcinoma clusters with 80% of the annotated ROIs. Once trained, the model reached a mIoU=0.74 and mDice=0.85 on a previously unseen evaluation set compound of the remaining 20% ROIs. The automatic method was then validated to attribute a score to each segmented CDH17+ cluster. DAB expressions were extracted for each cluster using color deconvolution and the best classification performances were obtained using the 90th percentile of the DAB distribution (Kappa score=0.74) compared to pathologist scoring. Finally, the whole workflow was launched on the entire image dataset to get predicted H-scores and compare them to the pathologist ones. The workflow reached a correlation R2=0.92 and a median absolute error=17.5 over all tumor locations. Our workflow reproduces the pathologist H-Score and shows a consistent performance on low- and high- range H-Scores (i.e. close to 0 and superior to 200), across all tumor locations of the study. This methodology fosters robustness and reproducibility of H-score quantification results for the CDH17 biomarker. Our workflow can straightforwardly be applied to other histological biomarkers, aiming to increase the rigorousness of its validation for future application in patient selection assays. Citation Format: Paul Klein, Thomas Le Meur, Roberto Fiorelli, Claudia Buehnemann, Aurelie Auguste, Sylvain Berlemont. A deep learning workflow to quantify the H-Score on immunohistochemistry images [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 193.

A core laboratory approach to large-scale radiomics and machine-learning prediction of DLBCL outcomes after first-line treatment using results from the phase III GOYA study.

e19042 Background: We are using the phase 3 GOYA study to retrospectively investigate using quantitative image texture features (i.e. radiomics, evaluating tumor heterogeneity) to improve the prognostic value of baseline FDG-PET. We report on the considerations for analysis, which is ongoing, and solutions to methodological challenges. Methods: GOYA, an open-label, randomized, phase 3 study, enrolled 1418 patients from 207 centers; 1334 had a baseline FDG-PET scan with detectable lesions. Image data with annotated regions of interest (ROIs) defined by qualified physicians were transferred to the Cuneo imaging core laboratory. Images were interpolated to a common 4x4x4 mm voxel size and ROI masks using 4 different definition methodologies were used to define image sub-volumes that contained the lesions. Large ROIs in normal liver were also defined. The 42 Haralick radiomic image texture features were computed using the PET oncology radiomics test suite. Results: A total of 9286 lesions were analyzed. Interpolation of images to a fixed size is challenging in this cohort due the large number of features to be computed (~1.5M). As a test, the lesions were differentiated from liver ROIs using 5 classifiers (e.g. support vector machine, random forest, etc) with an accuracy of 0.995–0.999. The mean (±SD) of the metabolically active tumor volume was 21.2±22.2 mL (range 0.2–146 mL) with a preponderance of smaller lesions as shown by the median of 15.0 mL. The skewed distribution had downstream effects on many of the quantitative (e.g. total lesion glycolysis) and radiomic image texture features. The majority of texture features were computable; 5 (Coarseness, Contrast, Busyness, Complexity, Texture Strength) required interpolation to a 2x2x2 mm voxel size, requiring extended computation time. FDG-PET images came from a variety of PET/CT scanners; thus, a unified image reconstruction protocol was not possible. Conclusions: Radiomics analysis of large phase 3 imaging trials is feasible. However, careful attention should be paid to the entire processing chain, including tests for non-computable features values due to highly skewed distributions and lesion size. Clinical trial information: NCT01287741.

Whole-Genome Analysis and HLA Genotyping of Enteropathy-Type T-Cell Lymphoma Reveals 2 Distinct Lymphoma Subtypes

Enteropathy-type T-cell lymphoma (ETL) is an aggressive extranodal T-cell non-Hodgkin lymphoma assumed to arise in the setting of celiac disease. To precisely define the genetic alterations underlying the pathogenesis of ETL, 30 ETL samples were profiled for genetic copy number alterations using high-resolution whole-genome tiling path array comparative genomic hybridization. To investigate the potential association of genetic alterations in ETL with celiac disease, HLA-DQB1 genotyping was performed. By array comparative genomic hybridization, 13 novel recurrent minimal regions of chromosomal alteration were identified on multiple chromosome arms. ETL is characterized by frequent complex gains of 9q31.3-qter (70% of cases), or by an almost mutually exclusive 2.5-megabase loss of 16q12.1 (23% of cases). Two distinct groups of ETL could be delineated morphologically and genetically: type 1 ETL is characterized by nonmonomorphic cytomorphology, CD56 negativity, and chromosomal gains of 1q and 5q. Type 1 ETL also appears to be linked pathogenetically to celiac disease, sharing genetic alterations and HLA-DQB1 genotype patterns with (refractory) celiac disease. Type 2 ETL shows monomorphic small- to medium-sized tumor cell morphology, frequently shows CD56 expression, MYC oncogene locus gain, and rare gains of chromosomes 1q and 5q. In contrast to type 1 ETL, type 2 ETL shows a HLA-DQB1 genotype pattern more resembling that of the normal Caucasian population. Contrary to current clinical classification, ETL comprises 2 morphologically, clinically, and genetically distinct lymphoma entities. In addition, type 2 ETL may not be associated with celiac disease.