Computer-assisted mitotic count using a deep learning–based algorithm improves interobserver reproducibility and accuracy

2021 ◽  
pp. 030098582110674
Author(s):  
Christof A. Bertram ◽  
Marc Aubreville ◽  
Taryn A. Donovan ◽  
Alexander Bartel ◽  
Frauke Wilm ◽  
...  
Keyword(s):  
High Performance ◽  
Region Of Interest ◽  
Ground Truth ◽  
Mitotic Count ◽  
Computer Assisted ◽  
Malignant Neoplasms ◽  
Mast Cell Tumors ◽  
Computer Assistance ◽  
Phosphohistone H3 ◽  
Cutaneous Mast Cell

The mitotic count (MC) is an important histological parameter for prognostication of malignant neoplasms. However, it has inter- and intraobserver discrepancies due to difficulties in selecting the region of interest (MC-ROI) and in identifying or classifying mitotic figures (MFs). Recent progress in the field of artificial intelligence has allowed the development of high-performance algorithms that may improve standardization of the MC. As algorithmic predictions are not flawless, computer-assisted review by pathologists may ensure reliability. In the present study, we compared partial (MC-ROI preselection) and full (additional visualization of MF candidates and display of algorithmic confidence values) computer-assisted MC analysis to the routine (unaided) MC analysis by 23 pathologists for whole-slide images of 50 canine cutaneous mast cell tumors (ccMCTs). Algorithmic predictions aimed to assist pathologists in detecting mitotic hotspot locations, reducing omission of MFs, and improving classification against imposters. The interobserver consistency for the MC significantly increased with computer assistance (interobserver correlation coefficient, ICC = 0.92) compared to the unaided approach (ICC = 0.70). Classification into prognostic stratifications had a higher accuracy with computer assistance. The algorithmically preselected hotspot MC-ROIs had a consistently higher MCs than the manually selected MC-ROIs. Compared to a ground truth (developed with immunohistochemistry for phosphohistone H3), pathologist performance in detecting individual MF was augmented when using computer assistance (F1-score of 0.68 increased to 0.79) with a reduction in false negatives by 38%. The results of this study demonstrate that computer assistance may lead to more reproducible and accurate MCs in ccMCTs.

scholarly journals Computer-Assisted Mitotic Count Using a Deep Learning-based Algorithm Improves Inter-Observer Reproducibility and Accuracy in canine cutaneous mast cell tumors

2021 ◽  
Author(s):  
Christof A Bertram ◽  
Marc Aubreville ◽  
Taryn A Donovan ◽  
Alexander Bartel ◽  
Frauke Wilm ◽  
...  
Keyword(s):  
Mast Cell ◽  
High Performance ◽  
Ground Truth ◽  
Mitotic Count ◽  
Computer Assisted ◽  
Malignant Neoplasms ◽  
Mast Cell Tumors ◽  
Computer Assistance ◽  
Phosphohistone H3 ◽  
Cutaneous Mast Cell

The mitotic count (MC) is an important histological parameter for prognostication of malignant neoplasms. However, it has inter- and intra-observer discrepancies due to difficulties in selecting the region of interest (MC-ROI) and in identifying/classifying mitotic figures (MFs). Recent progress in the field of artificial intelligence has allowed the development of high-performance algorithms that may improve standardization of the MC. As algorithmic predictions are not flawless, the computer-assisted review by pathologists may ensure reliability. In the present study we have compared partial (MC-ROI preselection) and full (additional visualization of MF candidate proposal and display of algorithmic confidence values) computer-assisted MC analysis to the routine (unaided) MC analysis by 23 pathologists for whole slide images of 50 canine cutaneous mast cell tumors (ccMCTs). Algorithmic predictions aimed to assist pathologists in detecting mitotic hotspot locations, reducing omission of MF and improving classification against imposters. The inter-observer consistency for the MC significantly increased with computer assistance (interobserver correlation coefficient, ICC = 0.92) compared to the unaided approach (ICC = 0.70). Classification into prognostic stratifications had a higher accuracy with computer assistance. The algorithmically preselected MC-ROIs had a consistently higher MCs than the manually selected MC-ROIs. Compared to a ground truth (developed with immunohistochemistry for phosphohistone H3), pathologist performance in detecting individual MF was augmented when using computer assistance (F1-score of 0.68 increased to 0.79) with a reduction in false negatives by 38%. The results of this study prove that computer assistance may lead to a more reproducible and accurate MCs in ccMCTs.

scholarly journals Deep learning algorithms out-perform veterinary pathologists in detecting the mitotically most active tumor region

2020 ◽  
Vol 10 (1) ◽  
Author(s):  
Marc Aubreville ◽  
Christof A. Bertram ◽  
Christian Marzahl ◽  
Corinne Gurtner ◽  
Martina Dettwiler ◽  
...  
Keyword(s):  
Deep Learning ◽  
Ground Truth ◽  
Mitotic Count ◽  
Mitotic Figure ◽  
Two Stage ◽  
Data Set ◽  
Area Selection ◽  
Tumor Region ◽  
Cutaneous Mast Cell ◽  
Mitotic Figures

Abstract Manual count of mitotic figures, which is determined in the tumor region with the highest mitotic activity, is a key parameter of most tumor grading schemes. It can be, however, strongly dependent on the area selection due to uneven mitotic figure distribution in the tumor section. We aimed to assess the question, how significantly the area selection could impact the mitotic count, which has a known high inter-rater disagreement. On a data set of 32 whole slide images of H&E-stained canine cutaneous mast cell tumor, fully annotated for mitotic figures, we asked eight veterinary pathologists (five board-certified, three in training) to select a field of interest for the mitotic count. To assess the potential difference on the mitotic count, we compared the mitotic count of the selected regions to the overall distribution on the slide. Additionally, we evaluated three deep learning-based methods for the assessment of highest mitotic density: In one approach, the model would directly try to predict the mitotic count for the presented image patches as a regression task. The second method aims at deriving a segmentation mask for mitotic figures, which is then used to obtain a mitotic density. Finally, we evaluated a two-stage object-detection pipeline based on state-of-the-art architectures to identify individual mitotic figures. We found that the predictions by all models were, on average, better than those of the experts. The two-stage object detector performed best and outperformed most of the human pathologists on the majority of tumor cases. The correlation between the predicted and the ground truth mitotic count was also best for this approach (0.963–0.979). Further, we found considerable differences in position selection between pathologists, which could partially explain the high variance that has been reported for the manual mitotic count. To achieve better inter-rater agreement, we propose to use a computer-based area selection for support of the pathologist in the manual mitotic count.

scholarly journals Computerized Calculation of Mitotic Count Distribution in Canine Cutaneous Mast Cell Tumor Sections: Mitotic Count Is Area Dependent

2019 ◽  
Vol 57 (2) ◽  
pp. 214-226 ◽  
Author(s):  
Christof A. Bertram ◽  
Marc Aubreville ◽  
Corinne Gurtner ◽  
Alexander Bartel ◽  
Sarah M. Corner ◽  
...  
Keyword(s):  
Mast Cell ◽  
High Power ◽  
Ground Truth ◽  
Mitotic Count ◽  
Automated Image Analysis ◽  
High Power Field ◽  
Borderline Cases ◽  
Area Selection ◽  
Cutaneous Mast Cell ◽  
Mitotic Figures

Mitotic count (MC) is an important element for grading canine cutaneous mast cell tumors (ccMCTs) and is determined in 10 consecutive high-power fields with the highest mitotic activity. However, there is variability in area selection between pathologists. In this study, the MC distribution and the effect of area selection on the MC were analyzed in ccMCTs. Two pathologists independently annotated all mitotic figures in whole-slide images of 28 ccMCTs (ground truth). Automated image analysis was used to examine the ground truth distribution of the MC throughout the tumor section area, which was compared with the manual MCs of 11 pathologists. Computerized analysis demonstrated high variability of the MC within different tumor areas. There were 6 MCTs with consistently low MCs (MC<7 in all tumor areas), 13 cases with mostly high MCs (MC ≥7 in ≥75% of 10 high-power field areas), and 9 borderline cases with variable MCs around 7, which is a cutoff value for ccMCT grading. There was inconsistency among pathologists in identifying the areas with the highest density of mitotic figures throughout the 3 ccMCT groups; only 51.9% of the counts were consistent with the highest 25% of the ground truth MC distribution. Regardless, there was substantial agreement between pathologists in detecting tumors with MC ≥7. Falsely low MCs below 7 mainly occurred in 4 of 9 borderline cases that had very few ground truth areas with MC ≥7. The findings of this study highlight the need to further standardize how to select the region of the tumor in which to determine the MC.

scholarly journals HSP32 and HSP90 Immunoexpression, in Relation to Kit Pattern, Grading, and Mitotic Count in Canine Cutaneous Mast Cell Tumors

2016 ◽  
Vol 54 (2) ◽  
pp. 222-225 ◽  
Author(s):  
M. Romanucci ◽  
M. Massimini ◽  
A. Ciccarelli ◽  
D. Malatesta ◽  
L. Bongiovanni ◽  
...  
Keyword(s):  
Mast Cell ◽  
Histological Grade ◽  
Tumor Grade ◽  
Mitotic Count ◽  
Immunohistochemical Expression ◽  
Mast Cell Tumors ◽  
Great Heterogeneity ◽  
Poorly Differentiated ◽  
Cutaneous Mast Cell

Literature data indicate heat shock protein (Hsp) 32 and 90 as potential molecular targets in canine neoplastic mast cells (MCs). However, their immunoexpression patterns in canine mast cell tumors (MCTs) have not been investigated. Thus, the aim of this study was to evaluate the immunohistochemical expression of Hsp32 and Hsp90 in 22 canine cutaneous MCTs, in relation to KIT immunolabeling pattern, histological grade, and mitotic count. All cases showed cytoplasmic labeling of Hsp90, variably associated with nuclear and/or membranous labeling. Relationships of Hsp90 or Hsp32 immunolabeling with KIT pattern, mitotic count, and tumor grade were not observed. However, the reduced Hsp32 immunoexpression observed in most grade III/high-grade MCTs suggests a tendency toward a loss of immunosignal in poorly differentiated MCs. The great heterogeneity in extent and distribution of Hsp90 immunoexpression among the different MCT cases may also partially explain the difficulties in predicting the in vivo biologic activity of Hsp90 inhibitors on canine MCTs.

scholarly journals Digital Lesions in Dogs: A Statistical Breed Analysis of 2912 Cases

2021 ◽  
Vol 8 (7) ◽  
pp. 136
Author(s):  
Julia Maria Grassinger ◽  
Andreas Floren ◽  
Tobias Müller ◽  
Argiñe Cerezo-Echevarria ◽  
Christoph Beitzinger ◽  
...  
Keyword(s):  
Bone Tumors ◽  
Soft Tissue Sarcomas ◽  
Benign Tumors ◽  
Malignant Neoplasms ◽  
Routine Diagnostics ◽  
Mast Cell Tumors ◽  
Neoplastic Processes ◽  
Study Population ◽  
Tumor Like Lesions ◽  
Labrador Retrievers

Breed predispositions to canine digital neoplasms are well known. However, there is currently no statistical analysis identifying the least affected breeds. To this end, 2912 canine amputated digits submitted from 2014–2019 to the Laboklin GmbH & Co. KG for routine diagnostics were statistically analyzed. The study population consisted of 155 different breeds (most common: 634 Mongrels, 411 Schnauzers, 197 Labrador Retrievers, 93 Golden Retrievers). Non-neoplastic processes were present in 1246 (43%), tumor-like lesions in 138 (5%), and neoplasms in 1528 cases (52%). Benign tumors (n = 335) were characterized by 217 subungual keratoacanthomas, 36 histiocytomas, 35 plasmacytomas, 16 papillomas, 12 melanocytomas, 9 sebaceous gland tumors, 6 lipomas, and 4 bone tumors. Malignant neoplasms (n = 1193) included 758 squamous cell carcinomas (SCC), 196 malignant melanomas (MM), 76 soft tissue sarcomas, 52 mast cell tumors, 37 non-specified sarcomas, 29 anaplastic neoplasms, 24 carcinomas, 20 bone tumors, and 1 histiocytic sarcoma. Predisposed breeds for SCC included the Schnauzer (log OR = 2.61), Briard (log OR = 1.78), Rottweiler (log OR = 1.54), Poodle (log OR = 1.40), and Dachshund (log OR = 1.30). Jack Russell Terriers (log OR = −2.95) were significantly less affected by SCC than Mongrels. Acral MM were significantly more frequent in Rottweilers (log OR = 1.88) and Labrador Retrievers (log OR = 1.09). In contrast, Dachshunds (log OR = −2.17), Jack Russell Terriers (log OR = −1.88), and Rhodesian Ridgebacks (log OR = −1.88) were rarely affected. This contrasted with the well-known predisposition of Dachshunds and Rhodesian Ridgebacks to oral and cutaneous melanocytic neoplasms. Further studies are needed to explain the underlying reasons for breed predisposition or “resistance” to the development of specific acral tumors and/or other sites.

scholarly journals Stereology in Grading and Prognosis of Canine Cutaneous Mast Cell Tumors

2021 ◽  
pp. 030098582098513
Author(s):  
Mafalda Casanova ◽  
Sandra Branco ◽  
Inês Berenguer Veiga ◽  
André Barros ◽  
Pedro Faísca
Keyword(s):  
Mast Cell ◽  
Clinical Outcome ◽  
Prognostic Value ◽  
Histological Grade ◽  
Intraobserver Variability ◽  
Low Grade ◽  
High Grade ◽  
Mast Cell Tumors ◽  
Cutaneous Mast Cell ◽  
Grade Iii

Canine cutaneous mast cell tumors (ccMCTs) are currently graded according to Patnaik and Kiupel grading schemes. The qualitative and semiquantitative parameters applied in these schemes may lead to inter- and intraobserver variability. This study investigates the prognostic value of volume-weighted mean nuclear volume ([Formula: see text]), a stereological estimation that provides information about nuclear size and its variability. [Formula: see text] of 55 ccMCTs was estimated using the “point-sampled intercept” method and compared with histological grade and clinical outcome. The clinical history of dogs treated with surgical excision alone was available for 30 ccMCTs. Statistical differences in [Formula: see text] were found between grade II ([Formula: see text]= 115 ± 29 µm3) and grade III ccMCTs ([Formula: see text]= 197 ± 63 µm3), as well as between low-grade ([Formula: see text]= 113 ± 28 µm3) and high-grade ccMCTs ([Formula: see text]= 184 ± 63 µm3). An optimal cutoff value of [Formula: see text] ≥ 150 µm3 and [Formula: see text] ≥ 140 µm3 was determined for grade III and high-grade ccMCTs, respectively. In terms of prognosis, [Formula: see text] was not able to predict the clinical outcome in 42% of the cases; however, cases with [Formula: see text]<125 µm3 had a favorable outcome. These results indicate that, despite having limited prognostic value when used as a solitary parameter, [Formula: see text] is highly reproducible and is associated with histological grade as well as with benign behavior.

Computer-assisted database indexing: the state-of-the-art

1994 ◽  
Vol 19 (1) ◽  
pp. 23-27
Author(s):  
Gail M. Hodge
Keyword(s):  
Data Base ◽  
Production System ◽  
State Of The Art ◽  
The State ◽  
Computer Support ◽  
Computer Assisted ◽  
Computer Assistance ◽  
The Future ◽  
Database Indexing ◽  
Operational Systems

Discusses the state-of-the-art in computer indexing, defines indexing and computer assistance, describes the reasons for renewed interest, identifies the types of computer support in use using selected operational systems, describes the integration of various computer supports in one data base production system, and speculates on the future.

Deep learning for automated detection and numbering of permanent teeth on panoramic images

2021 ◽  
Author(s):  
Mohamed Estai ◽  
Marc Tennant ◽  
Dieter Gebauer ◽  
Andrew Brostek ◽  
Janardhan Vignarajan ◽  
...  
Keyword(s):  
Deep Learning ◽  
High Performance ◽  
Detection System ◽  
Ground Truth ◽  
Automated Detection ◽  
Permanent Teeth ◽  
Automated Method ◽  
Step Procedure ◽  
Panoramic Images ◽  
Tooth Number

Objective: This study aimed to evaluate an automated detection system to detect and classify permanent teeth on orthopantomogram (OPG) images using convolutional neural networks (CNNs). Methods: In total, 591 digital OPGs were collected from patients older than 18 years. Three qualified dentists performed individual teeth labelling on images to generate the ground truth annotations. A three-step procedure, relying upon CNNs, was proposed for automated detection and classification of teeth. Firstly, U-Net, a type of CNN, performed preliminary segmentation of tooth regions or detecting regions of interest (ROIs) on panoramic images. Secondly, the Faster R-CNN, an advanced object detection architecture, identified each tooth within the ROI determined by the U-Net. Thirdly, VGG-16 architecture classified each tooth into 32 categories, and a tooth number was assigned. A total of 17,135 teeth cropped from 591 radiographs were used to train and validate the tooth detection and tooth numbering modules. 90% of OPG images were used for training, and the remaining 10% were used for validation. 10-folds cross-validation was performed for measuring the performance. The intersection over union (IoU), F1 score, precision, and recall (i.e. sensitivity) were used as metrics to evaluate the performance of resultant CNNs. Results: The ROI detection module had an IoU of 0.70. The tooth detection module achieved a recall of 0.99 and a precision of 0.99. The tooth numbering module had a recall, precision and F1 score of 0.98. Conclusion: The resultant automated method achieved high performance for automated tooth detection and numbering from OPG images. Deep learning can be helpful in the automatic filing of dental charts in general dentistry and forensic medicine.

Sign in / Sign up

Export Citation Format

Share Document