Semi-supervised recommender system for bone implant ratio recommendation

Millions of people suffer from bone loss and diseases each year, which give rise to bone-related medical treatments. Multi-material coatings composed of Silicon Substituted Hydroxyapatite (Si-HA) and Silver Substituted Hydroxyapatite (Ag-HA) potentiated its usage as an ideal bone implant material with favourable biological response and antibacterial properties, respectively. Each patient has different conditions (e.g. bone ages, bone density, position, and infection status) and specific bone implant requirement, thus requiring specific designs of Si-HA and Ag-HA bone implant material ratios. In this paper, a recommender system which took advantage of Wide Deep recommendation architecture and DenseNet architecture was provided and a semi-supervised training process was utilized to recommend appropriate bone implant material ratio (Si-HA:Ag-HA) for each patient to improve the bone disease treatment and recovery. The proposed recommendation system could achieve high accuracy ( 92.2% ) and efficiency (using only 10% of labeled data) on test dataset, which could be potentially utilized in future clinical usages and provide doctors with decision support to make better clinical decisions.