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

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.

Mitotic Figures—Normal, Atypical, and Imposters: A Guide to Identification

“…Imperfect interobserver and intraobserver reproducibility when obtaining the MC can result from (1) difficulty of differentiating MF from MLF (human error); (2) an inability to efficiently locate the region of highest mitotic density; and (3) impaired reproducibility of MF identification or classification while performing repetitive tasks. 7,11,13,20,22,27,28,34,35,40,45,51,57,65 With sufficient numbers of images, humans, and computer programs can be "trained" to differentiate MF, AMF, and MLF. Algorithms can assess the entire slide or several slides within a short period of time, are 100% reproducible, and (with good data sets and deep learning methods) they can be as accurate as pathologists.…”

“…Detection and classification results of a dual-stage deep learning-based algorithm for mitotic figures in digital images of canine cutaneous mast cell tumors. 7 Detections (first stage) are labeled by boxes. Classification (second stage) into MF or MLF is indicated by color: Numbers below the boxes (model scores) as well as colors of the boxes indicate how likely a structure is interpreted to be MF (dark green, score ≥0.5, Figs.…”

“…44 However, multiple studies have demonstrated disparity in reproducibility of the MC with interobserver agreement (k-values) ranging between 0.08 to 0.77 even if standard area is used. 13,20,22,27,28,45,51 Additional potential causes of inconsistency in the MC include variable identification of MF, 11,40,45,62 an inability to evaluate the complete tumor area, 7,11,34,57 and potentially pathologist fatigue from performing repetitive tasks. Disagreements of labeling structures as MF or mitotic-like figures (MLF) occurred with frequencies of 6% to 35%, 40 10% to 46%, 45 17%, 11,21,58,59 and 68% 62 and can be improved by establishing clear criteria for MF morphology.…”

“…Possible computer-assisted system for determination of mitotic count in whole-slide images of canine cutaneous mast cell tumors 7. …”

Abstract: Deep Learning Algorithms Out-perform Veterinary Pathologists in Detecting the Mitotically Most Active Tumor Region

“…The mitotic count has a known high inter-rater disagreement and is strongly dependent on the area selection due to uneven mitotic figure distribution. In our work [1], we assessed the question, how significantly the area selection could impact the mitotic count. On a data set of 32 cases of H&E-stained canine cutaneous mast cell tumor, fully annotated for mitotic figures, we asked eight veterinary pathologists to select a field of interest for the mitotic count, and retrieved the mitotic count for that area from the data set.…”