aetherSlide is an AI-powered digital pathology platform developed to address the growing challenges of aging populations, increasing cancer incidence, and the global shortage of pathologists. Designed as a “second pair of eyes” for pathologists, the platform digitizes and streamlines pathology workflows, improving diagnostic efficiency, accuracy, and consistency.

Built on a vendor-neutral design, aetherSlide supports leading whole-slide scanners from Hamamatsu, Philips, Roche, and other manufacturers, helping healthcare institutions avoid vendor lock-in. The platform supports DICOM standards and integrates seamlessly with third-party AI applications through the EMPAIA framework.

Pathologists can perform AI-assisted slide review, launch pre-analysis with a single click, or conduct real-time inference on selected regions of interest. Clinical studies have shown that AI assistance reduces review time by 31.5%, improves micrometastasis detection sensitivity from 81.94% to 95.83%, and increases isolated tumor cell detection sensitivity from 67.95% to 96.15%.

The platform has been successfully deployed in international healthcare institutions, including University Hospital Tübingen in Germany, and has obtained CE-IVD certification, demonstrating its commitment to quality, interoperability, and clinical excellence.

aetherAI Co., Ltd.

In 2022, Yeh's team published an ESCNN model in Nature Communications using Whole Slide Imaging (WSI) for gastric cancer lymph node metastasis detection. It achieves high accuracy without lesion-level annotations. Clinically, AI reduced review times by 31.5% (saving 50.7s/slide, or ~1,000 mins/year at Linkou Chang Gung). For Isolated Tumor Cells (ITCs), inter-observer variation dropped from 0.6388 to 0.1113.

Trained on 983 WSIs and tested on 1,156 nodes, it was validated via expert review and IHC staining. Federated Learning cross-validation at Kaohsiung Chang Gung yielded LN-level and Slide-level AUCs of 0.9831 and 0.9936, rivaling manual annotation-heavy models.

Micrometastasis and ITC AUCs reached 0.9940 and 0.9228 (Linkou), and 0.9868 and 0.9829 (Kaohsiung). AI boosted sensitivity for micrometastases (81.94% to 95.83%) and ITCs (67.95% to 96.15%), keeping macrometastasis near 100%. Physicians' MCC rose to 0.9795–0.9863, matching top experts. However, rare artifact false positives reduced specificity to 84.21%, which future data expansion aims to resolve.

aetherAI has established a quality management system for medical devices in accordance with the Medical Device Quality Management System Regulations (QMS), ISO 13485:2016, EU regulation 2017/746 (IVDR), and US Quality Management System Regulation (QMSR) requirements. Key quality control aspects of the production process include design and development, design change control, process control, and quality inspection.

• Design and development, and change control： Conduct design planning and input, output, verification, trial production, validation, transfer, and change control in accordance with the Design and Development Control Procedure.

• Process Control： Manufacture in accordance with the Process Control Procedure, and use the production records to verify process compliance and traceability.

• Quality Inspection： Incoming Quality Control (IQC), In-Process Quality Control (IPQC), and Final Quality Control (FQC) are performed in accordance with the Inspection and Testing Control Procedure. Records, as well as records (including test results), are maintained. Products are only released for shipment after passing inspection.

The above mechanisms ensure product traceability and consistency from design and manufacturing to testing and release, while maintaining stable quality, safety, and performance.

aetherAI fully integrates risk management into the product lifecycle. During the design and development planning and input stages, personnel from relevant departments are convened to form a risk assessment team to implement risk management, which shall be reviewed at least annually. Members of the risk assessment team must understand medical device risk management and possess relevant experience or qualified professional training to ensure the rigor and effectiveness of its assessments.

Product risk management procedures are implemented in accordance with ISO 14971:2019, including risk analysis, risk evaluation, risk measure management, and risk control. During risk analysis, potential hazards are identified based on the product characteristics of medical devices and in vitro diagnostic medical devices. The risk level (probability of occurrence of harm and its severity) of each hazard factor identified as posing a risk is analyzed and evaluated.

For each hazard factor, the assessment team formulates control measures and assigns responsible personnel for implementation. Control measures may include inherently safe design, protective measures, and information for safety. Following implementation, the risks must be reassessed to determine whether they have been reduced to an acceptable level. If the residual risk remains unacceptable, countermeasures must be re-formulated and residual risks shall be re-evaluated.

aetherAI has established the Post-Market Surveillance, Reporting, and Recall Procedure, conducts post-market surveillance of medical devices, and has a mechanism for collecting information on product usage in accordance with the procedure, to serve as a baseline for quality improvement.

For marketed nonconforming products, the Company initiates actions pursuant to the Corrective and Preventive Action Procedure. If a design change to a marketed product is required, it shall be executed in accordance with the Design and Development Control Procedure.

If analysis indicates that the product poses a potential risk to user safety and health, or that optimization is necessary, relevant departments shall be notified to take appropriate measures. Cause analysis shall be conducted, appropriate countermeasures shall be formulated and implemented, and implementation results shall be verified. These actions are to prevent recurrence and proactively mitigate future risks.

Huang SC, Chen CC, Lan J, et al. Deep neural network trained on gigapixel images improves lymph node metastasis detection in clinical settings. Nat Commun. 2022;13(1):3347. doi:10.1038/s41467-022-30746-1