Story by QuantumBasel · First published 3 November 2025
Moonlight AI is transforming cancer diagnostics by using AI to analyse blood smear slides, with the ultimate goal of achieving a faster and more accessible alternative to traditional molecular testing. In collaboration with QuantumBasel and the University of Basel, the team explored how quantum algorithms and AI could further enhance model performance, particularly in detecting circulating rare cells (CRCs).
The Challenge: A Bottleneck in Cancer Diagnostics
Haematological cancers impact more than 1.7 million people globally. Yet, fewer than 20% of eligible patients receive vital molecular testing due to high costs, complexity, and the need for specialised equipment. Currently, it takes more than 10 days to detect mutations using traditional methods like Next-Generation Sequencing (NGS), a delay that can critically impact treatment outcomes.
Moonlight AI envisioned a better way: by using AI to analyse standard blood smear slides. Their model aims to dramatically reduce turnaround time from more than 10 days to under one hour, allowing more patients the benefit of timely, targeted therapies.
But pushing AI to this level of sensitivity and scalability introduced a new challenge: computational complexity.
The Solution: Smarter AI, Ready for Quantum
That’s where QuantumBasel assisted. Together, the team co-developed two innovative machine learning approaches to detect circulating rare cells (CRCs), a critical biomarker in cancer detection:
- AI Optimisation: Improved performance by more than 10% (average AUC).
- AI Training Complexity Reduction: Reduced the number oftrained parameters by a factor of 5, with comparable performance.
- A quantum-ready pipeline with two submitted patent applications
This dual-track innovation strategy gives Moonlight AI a deployable, patent-pending classical solution today and a quantum-ready pipeline for tomorrow.
"Our collaboration with QuantumBasel has delivered results quickly and consistently. They treated us as thought partners and brought a practical mindset that truly sets them apart.” — Nicole Romano, Co-Founder and CTO, Moonlight AI
The Path Forward: From Prototype to Production
The success of the proof of concept led to the filing of patent applications to secure Moonlight AI’s IP. Now, possible next steps could focus on evolving the quantum algorithm from prototype to production, paving the way for real-world deployment.
By blending the best of classical and quantum computing, Moonlight AI and QuantumBasel are accelerating innovation in cancer diagnostics.
Additional Reading
Detection and classification of cell morphologies are carried out by sampling the patient’s blood and processing the sample through a DNA sequencer before analysing the results. A much more efficient way would be to use Multiple Instance Learning (MIL) AI tools. A key bottleneck in this methodology is the computational complexity and high number of trainable parameters in deep learning-based attention mechanisms used in multiple instance learning (MIL), which can hinder training efficiency.
To overcome this, this paper proposes a novel attention-based extreme learning machine (extreme MIL) approach that significantly reduces the number of trained parameters while achieving comparable performance to deep MIL models in rare cell detection. Deep MIL models train all parameters through backpropagation, while extreme MIL freezes most parameters in the attention mechanism, only training the final classification layer.
Our key improvements over the previous state-of-the-art include: higher-dimensional feature space in the aggregation step of the attention-based deep MIL model which led to a considerable improvement in performance (>10% in AUC) and robustness, domain-specialised pre-processing which yields substantial performance improvements across all models, and extreme MIL achieved comparable performance to deep MIL while reducing the number of trained parameters by a factor of 5. The paper can be found here.
