Department of Dermatology, Rabin Medical Center, Beilinson Hospital, Petach Tikva, Israel; Institute of Pathology, Sheba Medical Center, Ramat Gan, Israel; Department of Dermatology, Sheba Medical Center, Ramat Gan, Israel; Faculty of Medicine, Tel Aviv University, Israel.
Institute of Pathology, Sheba Medical Center, Ramat Gan, Israel.
Pathol Res Pract. 2024 Sep;261:155480. doi: 10.1016/j.prp.2024.155480. Epub 2024 Jul 21.
Cutaneous fungal infections are one of the most common skin conditions, hence, the burden of determining fungal elements upon microscopic examination with periodic acid-Schiff (PAS) and Gomori methenamine silver (GMS) stains, is very time consuming. Despite some morphological variability posing challenges to training artificial intelligence (AI)-based solutions, these structures are favored potential targets, enabling the recruitment of promising AI-based technologies. Herein, we present a novel AI solution for identifying skin fungal infections, potentially providing a decision support system for pathologists. Skin biopsies of patients diagnosed with a cutaneous fungal infection at the Sheba Medical Center, Israel between 2014 and 2023, were used. Samples were stained with PAS and GMS and digitized by the Philips IntelliSite scanner. DeePathology® STUDIO fungal elements were annotated and deemed as ground truth data after an overall revision by two specialist pathologists. Subsequently, they were used to create an AI-based solution, which has been further validated in other regions of interests. The study participants were divided into two cohorts. In the first cohort, the overall sensitivity of the algorithm was 0.8, specificity 0.97, F1 score 0.78; in the second, the overall sensitivity of the algorithm was 0.93, specificity 0.99, F1 score 0.95. The results obtained are encouraging as proof of concept for an AI-based fungi detection algorithm. DeePathology® STUDIO can be employed as a decision support system for pathologists when diagnosing a cutaneous fungal infection using PAS and GMS stains, thereby, saving time and money.
皮肤真菌感染是最常见的皮肤疾病之一，因此，通过过碘酸雪夫（PAS）和 Gomori 美蓝（GMS）染色在显微镜下确定真菌成分的负担非常耗时。尽管一些形态学的可变性对基于人工智能（AI）的解决方案的训练构成了挑战，但这些结构是有潜力的潜在目标，可以利用有前途的基于 AI 的技术。在此，我们提出了一种用于识别皮肤真菌感染的新型 AI 解决方案，可能为病理学家提供决策支持系统。使用了 2014 年至 2023 年间在以色列 Sheba 医疗中心诊断为皮肤真菌感染的患者的皮肤活检样本。样本用 PAS 和 GMS 染色，并由飞利浦 IntelliSite 扫描仪进行数字化。DeePathology® STUDIO 真菌元素经过两位专家病理学家的全面修订后被注释为真实数据。随后，它们被用于创建基于 AI 的解决方案，并在其他感兴趣的区域进一步验证。研究参与者分为两个队列。在第一个队列中，算法的整体敏感性为 0.8，特异性为 0.97，F1 得分为 0.78；在第二个队列中，算法的整体敏感性为 0.93，特异性为 0.99，F1 得分为 0.95。所获得的结果令人鼓舞，证明了基于 AI 的真菌检测算法的概念验证。当使用 PAS 和 GMS 染色诊断皮肤真菌感染时，DeePathology® STUDIO 可以作为病理学家的决策支持系统，从而节省时间和金钱。
