**1. Introduction**

The classical practice of medical is undergoing a transition as new scales of data production become increasingly common. This transition presents the field with major analytical challenges that necessitate new and creative ways of engaging. The use of machine learning is one of the most important developments supporting this transition, as its methods are ideally suited (and in fact benefit from) the massive scale of data collection that is increasingly becoming the norm. With the rapid growth in high-throughput technologies for genetics, proteomics, and other biological metrics, alongside the recent wide-spread adaptation of electronic health records [1, 2], more data than ever has become available to feed into a machine learning system. The classical practice of medicine is typified by giants such as Dr. William Osler [3] whose diagnostic acumen improved the lives of many. In the very near future (and, in many cases, the present), physicians and diagnosticians will work with more data than they could possibly interpret. Machine learning is one of many tools which will help alleviate this, helping to guide many diagnostic and therapeutic decisions made by the clinical team, and if implemented well, should support patients' overall health. This potential realization of "precision medicine" is based on the belief that each patient has unique characteristics which should be accounted for when treating them [4].

While precision medicine has already demonstrated major benefits in fields like pharmacology [5] and oncology [6–8], a number of potential applications remain in other medical fields. In this chapter, we will demonstrate this using spinal cord disease, specifically by examining its application to Degenerative Cervical Myelopathy (DCM). DCM is a condition when the bones and joints of the human neck (cervical spine) degenerate with age, causing a slow progressive 'squeeze' of the spinal cord. This progressive condition has a significant effect on patient quality of life. Symptoms include pain, numbness, dexterity loss, gait imbalance, and sphincter dysfunction [9, 10], with symptoms often not appearing until permanent damage has already occurred [11]. MRI scans are typically used as part of the diagnostic process, and demographic factors have also been shown to be effective in predicting DCM severity [12]. An additional challenge is that patients can exhibit the hallmarks of DCM without developing symptoms [13], suggesting that a wide range of factors may be contributing to the illness's severity. Despite all of this, research into precision medical approaches and diagnostics have been sorely lacking; to the best of our knowledge, only 4 published studies involving DCM (also referred to as Cervical Spondylotic Myelopathy) exist which utilize machine learning [7, 14–16], coming from only three different groups (including our own), and with only one utilizing MRI data [16].

In this chapter, we will discuss how we went about designing a machine learning process, focusing on considerations required for clinical data specifically. We first explore how data should be managed and stored, before moving into data preparation procedures. Finally, we move onto the design considerations for the machine learning model. We will focus on models made for diagnostic prediction, rather than outcome prediction; however, we intend this only as a first step in using machine learning to support patient care, with future work moving toward models that provide personalized therapeutic recommendations as well. Throughout this chapter we will apply the techniques being discussed to DCM to help contextualize them. Some preliminary results for the final resulting system will also be shown, as a 'proof-of-concept', using the CIFAR-10 dataset modified to replicate clinical circumstances. We hope that this will provide a road-map for future machine learning driven precision medicine projects to follow.
