Diabetes mellitus is a chronic disease with global implications, often taking form as the root cause of accelerated occurrence of other severe health ailments not limited to heart disorders, multiple organ failures and strokes. The World Health Organization reports a troubling statistic that 14% of adults worldwide are affected by some form or the other of diabetes. In the current global health scenario, it has become a matter of paramount importance to ensure early detection and effective diagnosis of diabetes to allow prompt treatment or, even more importantly, timely prevention. This paper proposes a novel quantum machine learning framework with a specialized, custom-designed quantum circuit that makes use of a novel randomizer based on cryptographic principles. Three custom-designed quantum circuits, along with two different cryptographic methods, namely the Hash-based Randomizer and the ChaCha20 Randomizer, were used in tandem to achieve unbiased selection of feature sets. The proposed framework not only improves existing work for diabetes detection, achieving a remarkable accuracy in prediction, but also paves the way for an effective method to determine the appropriate parameters of human biomarkers relevant to the detection of diabetes. In addition, the framework also exceeds the performance of existing machine learning algorithms in terms of prediction accuracy and the ability to handle complex non-linear data comprising 41 real-world biomarkers, with the possibility of enhanced scalability for adaption towards more complex physiological metrics. The accuracy obtained using the proposed framework falls in the range between 86 and 98% with the real-world dataset.
Naramparambath et al. (Fri,) studied this question.