As artificial intelligence continues to transform medical diagnostics, AI-powered microscopy is emerging as a powerful tool to enhance accuracy, efficiency, and accessibility in laboratory medicine. By combining advanced imaging, edge computing, and machine learning, new diagnostic platforms are enabling real-time analysis directly at the instrument, reducing dependence on cloud infrastructure while ensuring faster turnaround times and stronger data privacy.
In this interview with AI Spectrum, Tathagato Rai Dastidar, Founder and CEO of SigTuple, discusses the technological innovations behind the company’s AS76 AI-powered microscopy platform, an edge-AI system designed for automated slide analysis. He explains the engineering challenges of running high-throughput AI pipelines entirely on local hardware, the process of achieving diagnostic-grade accuracy through diverse clinical datasets, and how modular AI architectures enable continuous upgrades across multiple pathology applications. Dastidar also highlights how regulatory frameworks such as In Vitro Diagnostic Regulation (IVDR) are shaping the development of reliable and explainable AI systems, and how intelligent microscopy platforms could help bridge global shortages of laboratory specialists while expanding access to quality diagnostics.
What were the key technical challenges of running AS76 fully on local hardware?
The core challenge was building a deterministic, high-throughput pipeline where imaging, motion control, and GPU inference run in real time without cloud buffering. We had to optimise data flow from camera to GPU, manage thermals and sustained compute loads, and ensure robustness to real-world slide variability. In diagnostics, latency and predictability matter as much as accuracy.
How did you train and validate for diagnostic-grade accuracy?
We trained on diverse, multi-site datasets covering variations in staining, smear quality, and patient demographics. Ground truth was derived through expert-validated workflows. Validation is performed at both cell and case level, with cross-site testing to ensure generalisation. The system also flags uncertainty and quality issues to maintain clinical reliability.
How does edge AI reshape diagnostic workflows?
Edge AI delivers predictable turnaround time, stronger data privacy, and independence from unreliable bandwidth. It enables real-time analysis at the instrument itself, making advanced diagnostics feasible even in infrastructure-constrained settings.
How modular is the AI architecture?
AS76 is built as a modular platform. Core components like QC, detection, and normalisation are shared, while domain-specific modules handle leukocytes, morphology, marrow, fluids, and future applications. This allows additive upgrades without redesigning the system.
How did IVDR influence development?
IVDR enforces traceability, risk management, controlled updates, and clinically meaningful validation. We built structured datasets, strict version control, regression testing, and explainable outputs—such as visual evidence and quality flags—into the system from the outset.
How can AI microscopy bridge specialist shortages?
By automating routine screening and triaging abnormal cases, AI lets specialists focus on complex interpretation. With stable power, basic IT infrastructure, training, and service support, such systems can scale sustainably—even in remote regions.
