A novel approach to monitoring IVD devices via Machine Learning and Computer Vision

The use of artificial intelligence (AI) is becoming increasingly popular in machine vision systems, which are employed for the monitoring of operational processes associated with a range of devices. We present a novel approach for the monitoring of devices employed in vitro diagnostics (IVD). The focus is on the crucial phase of the tip grip, which is essential for the correct functioning of the device. We implemented two AI algorithms – a support vector machine and a convolutional neural network – for the verification of the tip presence and size. Both models were trained using k -fold cross-validation. The validation phase yielded excellent results confirming the efficacy of the proposed methodology for this task. In addition, the explainability of the convolutional neural network model was analyzed in order to gain deeper insights into its decision-making process. This research contributes to the safety and effectiveness of IVD devices, paving the way for further developments in AI applications for laboratory medicine. • A novel machine vision approach enables real-time tip presence/size checks in IVDs. • Support Vector Machines reach 99.3% on unseen data; CNN achieves 100%. • Explainability reveals CNNs focus on key tip regions, enhancing transparency. • The solution improves operational safety and efficiency in laboratory diagnostics. • Future work: expand IVD monitoring and explore Transformer-based models.