Secondary Logo

Journal Logo

Research Highlights

Barisoni, Laura1; Luo, Xunrong2

doi: 10.1097/TP.0000000000002972
In View: Research Highlights
Free

1 Department of Pathology, Duke University School of Medicine, Durham, NC.

2 Division of Nephrology, Department of Medicine, Duke University School of Medicine, Durham, NC.

Received 10 September 2019.

Accepted 10 September 2019.

The author declares no funding or conflicts of interest.

Back to Top | Article Outline

Clinical-grade Computational Pathology Using Weakly Supervised Deep Learning on Whole Slide Images

Campanella G, Hanna MG, Geneslaw L, et al. Nat Med. 2019; 25:1301–1309.

Computational image analysis is an evolving field that relies on digital whole slide imaging and computer algorithms to create decision support tools for pathologists, clinicians, researchers, and patients. The immediate goal is to enhance the efficiency and accuracy of morphologic assessment, while enabling generalizability, which should result in better global patient care. While this science has been operating for some time in radiology, it is in its infancy in pathology. Generating such computer algorithms has so far relied on small, supervised (manually annotated) datasets for deep machine learning and training. The application of such algorithms to large-scale unsupervised datasets, however, is often problematic due to wide variances of clinical samples typically not captured by small training datasets. In addressing this challenge, Campanella et al1 presented a new framework for training at a very large scale based on “multiple instance learning,” aiming to train a classification model in a weakly supervised manner. This process employs information from anatomic pathology laboratory information system and/or electronic health records, and involves the following several steps: (1) tiling of the entire slide image; (2) training at the tile level and ranking the tiles according to their probability of being positive; (3) integrating information across the entire slide using a recurrent neural network; and (4) correlating with or reporting the final classification result. To test the validity of this approach, the authors used a binary system (presence or absence of tumors on the whole slide images), with the goal of achieving 100% sensitivity while maximizing the area under the curve of the receiver operating characteristic curve. Achieving 100% sensitivity would allow such an algorithm to be used as an effective digital screening tool, minimizing the work load of pathologists while not missing any potential positive cases. The authors tested 3 large datasets involving prostate cancer, basal cell carcinoma, and breast cancer metastasis to axillary lymph nodes, totaling 44 732 whole slide images from 15 187 patients, all without any data curation, were tested. To further address generalizability, the datasets were provided by different sites with the use of different slide scanners. This framework achieved an area under the curve of >0.98 for all cancer types, excluding 65%–75% of negative slides while retaining 100% sensitivity.

Such an approach could be applied to transplant clinical care. For example, optimal caring for organ recipients is often critically dependent on our ability to timely detect immunological activities and accurately discern nature, mechanisms, and etiology of structural changes from implant, for cause or protocol biopsies.2 Implementing such a computational decision support system would therefore significantly enhance our efficiency and precision in our clinical practices. Thrre of the following advances of such a system would be highly advantageous for its application in transplantation: (1) deep learning annotation and classification of tissue structural changes predictive of mechanisms and outcome; (2) algorithms that efficiently assess continuous variables rather than binary outputs; and (3) the ability to report the final classification also as continuous, quantifiable outputs. The framework demonstrated here exhibits the ability to train accurate classification models at an unprecedently large scale, providing a solid basis for future improvements, and development of computational decision support systems for practical clinical utilities.

1. Campanella G, Hanna MG, Geneslaw L, et al. Clinical-grade computational pathology using weakly supervised deep learning on whole slide images.Nat Med2019251301–1309

2. Loupy A, Haas M, Solez K, et al. The banff 2015 kidney meeting report: current challenges in rejection classification and prospects for adopting molecular pathology.Am J Transplant20171728–41

Back to Top | Article Outline

An Anti-CD3 Antibody, Teplizumab, in Relatives at Risk for Type 1 Diabetes

Herold KC, Bundy BN, Long SA, et al; Type 1 Diabetes TrialNet Study Group. N Engl J Med. 2019;381:603–613

CD3 is a protein complex consisting of 4 subunits acting as a co-receptor for T cell receptors to promote CD4 and CD8 T cell activation. Consequently, targeting CD3 has long been thought as a strategy for modulating T cell–mediated immune responses in transplantation and autoimmunity. Multiple mechanisms of action have been described for anti-CD3 involving induction of apoptosis of activated T cells and/or expansion of adaptive regulator T cells. An early anti-CD3 monoclonal antibody OKT3 has been used for various clinical indications including transplantation. However, this early form of anti-CD3 caused severe cytokine-release syndrome and induced host production of human anti-xenogeneic antibodies owing to its murine origin, significantly limiting it clinical applications. These untoward side effects have been successfully eliminated by subsequent iterations of this antibody that (1) are non-Fc binding and (2) are humanized. One such antibody is teplizumab. Treatment with teplizumab in patients with recent onset type 1 diabetes mellitus (T1DM) has been shown to reduce the rate of beta-cell loss.1 However, the immunomodulatory efficacy of this antibody in a prediabetic population was not known. In the current study, Herold et al2 conducted a phase 2 randomized placebo-controlled, double-blinded trial on the effect of teplizumab in a prediabetic high-risk population. In this study, a total of 76 nondiabetic relatives of known T1DM patients were enrolled. To qualify, the subject had at least 2 T1DM-related autoantibodies and an abnormal oral glucose tolerance test. The 14-day treatment course of teplizumab consisted of an escalating dose from 51 to 826 μg/m2 of body-surface area in the first 4 days followed by the stable dose at 826 μg/m2 for the remaining 10 days. This single course of teplizumab resulted in a remarkable delay in the progression of T1DM at 2 years following randomization, with 43% diagnosed with T1DM in the teplizumab group versus 72% in the placebo groups. Importantly, an increase of CD8+KLRG1+TIGIT+EOMES+ cells known to be associated with T cell unresponsiveness was observed in teplizumab-treated subjects, suggesting a possible mechanism of protection by teplizumab in the T1DM at-risk population. With this new information now available, it would be highly interesting to reexamine the role of an anti-CD3-based therapy in transplantation. The use of anti-CD3 monoclonal antibody for transplantation of T1DM patients makes the most logical sense, particularly if there is a concern of recurrent autoimmunity affecting the function of the transplanted organs/cells.3 Anti-CD3 in the presence of alloantigens may also promote donor-specific activation-induced cell death and donor-specific regulatory T cells, therefore functioning as a protolerogenic agent. For both transplantation and autoimmunity, further enhancement of teplizumab efficacy with repetitive dosing, extending findings from the current study, in settings of an acceptable adverse effect profile should now be tested.

1. Herold KC, Gitelman SE, Ehlers MR, et al; AbATE Study TeamTeplizumab (anti-CD3 mAb) treatment preserves C-peptide responses in patients with new-onset type 1 diabetes in a randomized controlled trial: metabolic and immunologic features at baseline identify a subgroup of responders.Diabetes2013623766–3774

2. Herold KC, Bundy BN, Long SA, et al; Type 1 Diabetes TrialNet Study GroupAn anti-CD3 antibody, teplizumab, in relatives at risk for type 1 diabetes.N Engl J Med2019381603–613

3. Bellin MD, Barton FB, Heitman A, et al. Potent induction immunotherapy promotes long-term insulin independence after islet transplantation in type 1 diabetes.Am J Transplant2012121576–1583

Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.