Erratum: Automatic textual description of colorectal polyp features: explainable artificial intelligence.

[This corrects the article DOI: 10.1055/a-2071-6652.].

Automatic textual description of colorectal polyp features: explainable artificial intelligence.

Computer-aided diagnosis systems (CADx) can improve colorectal polyp (CRP) optical diagnosis. For integration into clinical practice, better understanding of artificial intelligence (AI) by endoscopists is needed. We aimed to develop an explainable AI CADx capable of automatically generating textual descriptions of CRPs. For training and testing of this CADx, textual descriptions of CRP size and features according to the Blue Light Imaging (BLI) Adenoma Serrated International Classification (BASIC) were used, describing CRP surface, pit pattern, and vessels. CADx was tested using BLI images of 55 CRPs. Reference descriptions with agreement by at least five out of six expert endoscopists were used as gold standard. CADx performance was analyzed by calculating agreement between the CADx generated descriptions and reference descriptions. CADx development for automatic textual description of CRP features succeeded. Gwet's AC1 values comparing the reference and generated descriptions per CRP feature were: size 0.496, surface-mucus 0.930, surface-regularity 0.926, surface-depression 0.940, pits-features 0.921, pits-type 0.957, pits-distribution 0.167, and vessels 0.778. CADx performance differed per CRP feature and was particularly high for surface descriptors while size and pits-distribution description need improvement. Explainable AI can help comprehend reasoning behind CADx diagnoses and therefore facilitate integration into clinical practice and increase trust in AI.

Automatic image and text-based description for colorectal polyps using BASIC classification.

Colorectal polyps (CRP) are precursor lesions of colorectal cancer (CRC). Correct identification of CRPs during in-vivo colonoscopy is supported by the endoscopist's expertise and medical classification models. A recent developed classification model is the Blue light imaging Adenoma Serrated International Classification (BASIC) which describes the differences between non-neoplastic and neoplastic lesions acquired with blue light imaging (BLI). Computer-aided detection (CADe) and diagnosis (CADx) systems are efficient at visually assisting with medical decisions but fall short at translating decisions into relevant clinical information. The communication between machine and medical expert is of crucial importance to improve diagnosis of CRP during in-vivo procedures. In this work, the combination of a polyp image classification model and a language model is proposed to develop a CADx system that automatically generates text comparable to the human language employed by endoscopists. The developed system generates equivalent sentences as the human-reference and describes CRP images acquired with white light (WL), blue light imaging (BLI) and linked color imaging (LCI). An image feature encoder and a BERT module are employed to build the AI model and an external test set is used to evaluate the results and compute the linguistic metrics. The experimental results show the construction of complete sentences with an established metric scores of BLEU-1 = 0.67, ROUGE-L = 0.83 and METEOR = 0.50. The developed CADx system for automatic CRP image captioning facilitates future advances towards automatic reporting and may help reduce time-consuming histology assessment.

Diagnosis of Pituitary Adenoma Biopsies by Ultrahigh Resolution Optical Coherence Tomography Using Neuronal Networks.

Objective: Despite advancements of intraoperative visualization, the difficulty to visually distinguish adenoma from adjacent pituitary gland due to textural similarities may lead to incomplete adenoma resection or impairment of pituitary function. The aim of this study was to investigate optical coherence tomography (OCT) imaging in combination with a convolutional neural network (CNN) for objectively identify pituitary adenoma tissue in an ex vivo setting. Methods: A prospective study was conducted to train and test a CNN algorithm to identify pituitary adenoma tissue in OCT images of adenoma and adjacent pituitary gland samples. From each sample, 500 slices of adjacent cross-sectional OCT images were used for CNN classification. Results: OCT data acquisition was feasible in 19/20 (95%) patients. The 16.000 OCT slices of 16/19 of cases were employed for creating a trained CNN algorithm (70% for training, 15% for validating the classifier). Thereafter, the classifier was tested on the paired samples of three patients (3.000 slices). The CNN correctly predicted adenoma in the 3 adenoma samples (98%, 100% and 84% respectively), and correctly predicted gland and transition zone in the 3 samples from the adjacent pituitary gland. Conclusion: Trained convolutional neural network computing has the potential for fast and objective identification of pituitary adenoma tissue in OCT images with high sensitivity ex vivo. However, further investigation with larger number of samples is required.

Algorithm combining virtual chromoendoscopy features for colorectal polyp classification.

Background and study aims Colonoscopy is considered the gold standard for decreasing colorectal cancer incidence and mortality. Optical diagnosis of colorectal polyps (CRPs) is an ongoing challenge in clinical colonoscopy and its accuracy among endoscopists varies widely. Computer-aided diagnosis (CAD) for CRP characterization may help to improve this accuracy. In this study, we investigated the diagnostic accuracy of a novel algorithm for polyp malignancy classification by exploiting the complementary information revealed by three specific modalities. Methods We developed a CAD algorithm for CRP characterization based on high-definition, non-magnified white light (HDWL), Blue light imaging (BLI) and linked color imaging (LCI) still images from routine exams. All CRPs were collected prospectively and classified into benign or premalignant using histopathology as gold standard. Images and data were used to train the CAD algorithm using triplet network architecture. Our training dataset was validated using a threefold cross validation. Results In total 609 colonoscopy images of 203 CRPs of 154 consecutive patients were collected. A total of 174 CRPs were found to be premalignant and 29 were benign. Combining the triplet network features with all three image enhancement modalities resulted in an accuracy of 90.6 %, 89.7 % sensitivity, 96.6 % specificity, a positive predictive value of 99.4 %, and a negative predictive value of 60.9 % for CRP malignancy classification. The classification time for our CAD algorithm was approximately 90 ms per image. Conclusions Our novel approach and algorithm for CRP classification differentiates accurately between benign and premalignant polyps in non-magnified endoscopic images. This is the first algorithm combining three optical modalities (HDWL/BLI/LCI) exploiting the triplet network approach.

Prospective development and validation of a volumetric laser endomicroscopy computer algorithm for detection of Barrett's neoplasia.

BACKGROUND AND AIMS: Volumetric laser endomicroscopy (VLE) is an advanced imaging modality used to detect Barrett's esophagus (BE) dysplasia. However, real-time interpretation of VLE scans is complex and time-consuming. Computer-aided detection (CAD) may help in the process of VLE image interpretation. Our aim was to train and validate a CAD algorithm for VLE-based detection of BE neoplasia. METHODS: The multicenter, VLE PREDICT study, prospectively enrolled 47 patients with BE. In total, 229 nondysplastic BE and 89 neoplastic (high-grade dysplasia/esophageal adenocarcinoma) targets were laser marked under VLE guidance and subsequently underwent a biopsy for histologic diagnosis. Deep convolutional neural networks were used to construct a CAD algorithm for differentiation between nondysplastic and neoplastic BE tissue. The CAD algorithm was trained on a set consisting of the first 22 patients (134 nondysplastic BE and 38 neoplastic targets) and validated on a separate test set from patients 23 to 47 (95 nondysplastic BE and 51 neoplastic targets). The performance of the algorithm was benchmarked against the performance of 10 VLE experts. RESULTS: Using the training set to construct the algorithm resulted in an accuracy of 92%, sensitivity of 95%, and specificity of 92%. When performance was assessed on the test set, accuracy, sensitivity, and specificity were 85%, 91%, and 82%, respectively. The algorithm outperformed all 10 VLE experts, who demonstrated an overall accuracy of 77%, sensitivity of 70%, and specificity of 81%. CONCLUSIONS: We developed, validated, and benchmarked a VLE CAD algorithm for detection of BE neoplasia using prospectively collected and biopsy-correlated VLE targets. The algorithm detected neoplasia with high accuracy and outperformed 10 VLE experts. (The Netherlands National Trials Registry (NTR) number: NTR 6728.).

Optical diagnosis of colorectal polyp images using a newly developed computer-aided diagnosis system (CADx) compared with intuitive optical diagnosis.

BACKGROUND: Optical diagnosis of colorectal polyps remains challenging. Image-enhancement techniques such as narrow-band imaging and blue-light imaging (BLI) can improve optical diagnosis. We developed and prospectively validated a computer-aided diagnosis system (CADx) using high-definition white-light (HDWL) and BLI images, and compared the system with the optical diagnosis of expert and novice endoscopists. METHODS: CADx characterized colorectal polyps by exploiting artificial neural networks. Six experts and 13 novices optically diagnosed 60 colorectal polyps based on intuition. After 4 weeks, the same set of images was permuted and optically diagnosed using the BLI Adenoma Serrated International Classification (BASIC). RESULTS: CADx had a diagnostic accuracy of 88.3 % using HDWL images and 86.7 % using BLI images. The overall diagnostic accuracy combining HDWL and BLI (multimodal imaging) was 95.0 %, which was significantly higher than that of experts (81.7 %, P = 0.03) and novices (66.7 %, P < 0.001). Sensitivity was also higher for CADx (95.6 % vs. 61.1 % and 55.4 %), whereas specificity was higher for experts compared with CADx and novices (95.6 % vs. 93.3 % and 93.2 %). For endoscopists, diagnostic accuracy did not increase when using BASIC, either for experts (intuition 79.5 % vs. BASIC 81.7 %, P = 0.14) or for novices (intuition 66.7 % vs. BASIC 66.5 %, P = 0.95). CONCLUSION: CADx had a significantly higher diagnostic accuracy than experts and novices for the optical diagnosis of colorectal polyps. Multimodal imaging, incorporating both HDWL and BLI, improved the diagnostic accuracy of CADx. BASIC did not increase the diagnostic accuracy of endoscopists compared with intuitive optical diagnosis.

Hyperspectral imaging for colon cancer classification in surgical specimens: towards optical biopsy during image-guided surgery.

The main curative treatment for localized colon cancer is surgical resection. However when tumor residuals are left positive margins are found during the histological examinations and additional treatment is needed to inhibit recurrence. Hyperspectral imaging (HSI) can offer non-invasive surgical guidance with the potential of optimizing the surgical effectiveness. In this paper we investigate the capability of HSI for automated colon cancer detection in six ex-vivo specimens employing a spectral-spatial patch-based classification approach. The results demonstrate the feasibility in assessing the benign and malignant boundaries of the lesion with a sensitivity of 0.88 and specificity of 0.78. The results are compared with the state-of-the-art deep learning based approaches. The method with a new hybrid CNN outperforms the state-of the-art approaches (0.74 vs. 0.82 AUC). This study paves the way for further investigation towards improving surgical outcomes with HSI.

Towards ultrahigh resolution OCT based endoscopical pituitary gland and adenoma screening: a performance parameter evaluation.

Ultrahigh resolution optical coherence tomography (UHR-OCT) for differentiating pituitary gland versus adenoma tissue has been investigated for the first time, indicating more than 80% accuracy. For biomarker identification, OCT images of paraffin embedded tissue are correlated to histopathological slices. The identified biomarkers are verified on fresh biopsies. Additionally, an approach, based on resolution modified UHR-OCT ex vivo data, investigating optical performance parameters for the realization in an in vivo endoscope is presented and evaluated. The identified morphological features-cell groups with reticulin framework-detectable with UHR-OCT showcase a promising differentiation ability, encouraging endoscopic OCT probe development for in vivo application.

Cortical and trabecular bone analysis of professional dancers using 3D-DXA: a case-control study.

Given the lack of relevant data, the aim of this study was to examine femur cortical and trabecular bone in female and male professional ballet dancers. 40 professional ballet dancers and 40 sex- and age-matched non-exercising controls volunteered. Femoral bone density was scanned by dual-energy X-ray absorptiometry (DXA) scan. A 3D-DXA software was used to analyse trabecular and cortical bone. Anthropometry, maturation (Tanner staging), menstrual parameters (age at menarche and primary amenorrhea), energy availability and nutritional analysis (3-day record) were also assessed.Compared to non-exercising participants, dancers exhibited significantly higher volumetric density for integral, cortical and trabecular bone, and thicker cortex at the femur. Ballet dancers demonstrated lower body weight compared to controls (p < 0.01). Female dancers had their menarche later than controls, and the prevalence of primary amenorrhea were significantly higher in dancers than controls (p < 0.01). Dancer's energy availability was below the normal range (<30 kcal/kgFFM/day). Despite the presence of certain osteoporosis risk factors such as low energy availability, primary amenorrhoea and lower body weight, professional ballet dancers revealed higher bone density for both cortical and trabecular bone compartments compared to controls.

Cortical and Trabecular Bone Analysis of Patients With High Bone Mass From the Barcelona Osteoporosis Cohort Using 3-Dimensional Dual-Energy X-ray Absorptiometry: A Case-Control Study.

High bone mass (HBM), a rare phenotype, can be detected by dual-energy X-ray absorptiometry (DXA) scanning. Measurements with peripheral quantitative computed tomography at the tibia have found increased trabecular bone mineral density and changes in cortical bone density and structure, all of which lead to increased bone strength. However, no studies on cortical and trabecular bone have been performed at the femur. The recently developed 3-dimensional (3D)-DXA software algorithm quantifies the trabecular and cortical volumetric bone mineral density (vBMD) and the anatomical distribution of cortical thickness using routine hip DXA scans. We analyzed the femurs of 15 women with HBM and 15 controls from the Barcelona Osteoporosis (BARCOS) cohort using the 3D-DXA technique. The mean vBMD of proximal femur was 29.7% higher in HBM cases than in controls for the integral bone, 41.3% higher for the trabecular bone, and 7.3% higher for the cortical bone (p < 0.001). No differences in bone size were detected between cases and controls. Patients with HBM had a thicker cortex and higher trabecular and cortical vBMDs, as measured by 3D-DXA at the femur and compared to controls; bone size was similar in both groups. To the best of our knowledge, this is the first description of trabecular and cortical characteristics of the hip in patients with HBM.

3D-DXA: Assessing the Femoral Shape, the Trabecular Macrostructure and the Cortex in 3D from DXA images.

The 3D distribution of the cortical and trabecular bone mass in the proximal femur is a critical component in determining fracture resistance that is not taken into account in clinical routine Dual-energy X-ray Absorptiometry (DXA) examination. In this paper, a statistical shape and appearance model together with a 3D-2D registration approach are used to model the femoral shape and bone density distribution in 3D from an anteroposterior DXA projection. A model-based algorithm is subsequently used to segment the cortex and build a 3D map of the cortical thickness and density. Measurements characterising the geometry and density distribution were computed for various regions of interest in both cortical and trabecular compartments. Models and measurements provided by the "3D-DXA" software algorithm were evaluated using a database of 157 study subjects, by comparing 3D-DXA analyses (using DXA scanners from three manufacturers) with measurements performed by Quantitative Computed Tomography (QCT). The mean point-to-surface distance between 3D-DXA and QCT femoral shapes was 0.93 mm. The mean absolute error between cortical thickness and density estimates measured by 3D-DXA and QCT was 0.33 mm and 72 mg/cm3. Correlation coefficients (R) between the 3D-DXA and QCT measurements were 0.86, 0.93, and 0.95 for the volumetric bone mineral density at the trabecular, cortical, and integral compartments respectively, and 0.91 for the mean cortical thickness. 3D-DXA provides a detailed analysis of the proximal femur, including a separate assessment of the cortical layer and trabecular macrostructure, which could potentially improve osteoporosis management while maintaining DXA as the standard routine modality.