The histological distinction of lung neuroendocrine carcinoma, including small cell lung carcinoma (SCLC), large cell neuroendocrine carcinoma (LCNEC) and atypical carcinoid (AC), can be challenging in some cases, while bearing prognostic and therapeutic significance. To assist pathologists with the differentiation of histologic subtyping, we applied a deep learning classifier equipped with a convolutional neural network (CNN) to recognize lung neuroendocrine neoplasms. Slides of primary lung SCLC, LCNEC and AC were obtained from the Laboratory of Clinical and Experimental Pathology (University Hospital Nice, France). Three thoracic pathologists blindly established gold standard diagnoses. The HALO-AI module (Indica Labs, UK) trained with 18,752 image tiles extracted from 60 slides (SCLC = 20, LCNEC = 20, AC = 20 cases) was then tested on 90 slides (SCLC = 26, LCNEC = 22, AC = 13 and combined SCLC with LCNEC = 4 cases; NSCLC = 25 cases) by F1-score and accuracy. A HALO-AI correct area distribution (AD) cutoff of 50% or more was required to credit the CNN with the correct diagnosis. The tumor maps were false colored and displayed side by side to original hematoxylin and eosin slides with superimposed pathologist annotations. The trained HALO-AI yielded a mean F1-score of 0.99 (95% CI, 0.939–0.999) on the testing set. Our CNN model, providing further larger validation, has the potential to work side by side with the pathologist to accurately differentiate between the different lung neuroendocrine carcinoma in challenging cases.

PurposeImmune checkpoint inhibitors (ICIs) are now one of the standards of care for patients with lung cancer and have greatly improved both progression-free and overall survival, although <20% of the patients respond to the treatment, and some face acute adverse events. Although a few predictive biomarkers have integrated the clinical workflow, they require additional modalities on top of whole-slide images and lack efficiency or robustness. In this work, we propose a biomarker of immunotherapy outcome derived solely from the analysis of histology slides.ApproachWe develop a three-step framework, combining contrastive learning and nonparametric clustering to distinguish tissue patterns within the slides, before exploiting the adjacencies of previously defined regions to derive features and train a proportional hazards model for survival analysis. We test our approach on an in-house dataset of 193 patients from 5 medical centers and compare it with the gold standard tumor proportion score (TPS) biomarker.ResultsOn a fivefold cross-validation (CV) of the entire dataset, the whole-slide image-based survival analysis for patients treated with immunotherapy (WhARIO) features are able to separate a low- and a high-risk group of patients with a hazard ratio (HR) of 2.29 (CI95=1.48 to 3.56), whereas the TPS 1% reference threshold only reaches a HR of 1.81 (CI95=1.21 to 2.69). Combining the two yields a higher HR of 2.60 (CI95=1.72 to 3.94). Additional experiments on the same dataset, where one out of five centers is excluded from the CV and used as a test set, confirm these trends.ConclusionsOur uniquely designed WhARIO features are an efficient predictor of survival for lung cancer patients who received ICI treatment. We achieve similar performance to the current gold standard biomarker, without the need to access other imaging modalities, and show that both can be used together to reach even better results.

With the growing standardization of Whole Slide Images (WSIs), deep learning algorithms have shown promising results for the automated classification and localization of tumors. Yet, it is often difficult to train such algorithms, as they usually require careful detailed annotations from expert pathologists, which are tedious to produce. This is why in general only slide-level labels are accessible while annotations of small regions (or tiles) are limited. With only slide-level information, it is difficult to obtain accurate predictions of the localization of pathological tissues inside a slide, despite reaching good slide-level classification. Besides, existing algorithms show limited consistency between slide- and tile-level predictions, leading to difficult interpretation in case of healthy tissue. Using the attention-based multiple instance learning framework, we propose to combine slide-level labels on all slides with tile-level labels on a small fraction (e.g. 20%) of slides within a histology dataset to improve both classification and localization performances. With this mixed supervision of slides, we aim to enforce a better consistency between slide- and tile-level labels. To this end, we introduce an attention-based loss function to further guide the model’s attention on discriminative regions inside tumorous slides, and display an equal attention among all tiles of normal slides. On the Camelyon16 dataset, we reached precision and recall scores as high as 0.99 and 0.85 respectively with an AUC of 0.93 on the competition test set, using only 50% of the slides with tile-level annotations in the training set. Experiments using various proportions of fully annotated slides in the training set show promising results for an improved localization of tumors and classification of slides. In this work, we showed that using a limited amount of fully annotated slides we can improve both the classification and localization performances of an attention-based deep learning model. This increased consistency and performances should help pathologists to better interpret the algorithm output and to focus on suspicious regions in probable tumorous slides. Citation Format: Paul Tourniaire, Marius Ilie, Paul Hofman, Nicholas Ayache, Hervé Delingette. Mixed supervision to improve the classification and localization: Coherence of tumors in histological slides [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2022; 2022 Apr 8-13. Philadelphia (PA): AACR; Cancer Res 2022;82(12_Suppl):Abstract nr 461.