Data Mining Applications in Clinical Implementation and Healthcare: Highlights of Findings, Interventions, and Future Directions

Medical applications of data mining facilitate developments of clinical implementations in healthcare. The editors’ choice of this month make their own contributions towards healthcare industry using data mining techniques. The first editors’ choice, “EEG-based mild depressive detection using feature selection methods and classifiers [1],” presents a mild depression detection system using feature selection methods and data mining classifiers. First, electroencephalography (EEG) data were collected from both mild depressive patients and normal controls. After that, machine learning algorithms were applied to discriminate these two groups. Finally, feature selection methods were incorporated to identify discriminative electrodes and features. Li et al. demonstrated that the proposed system achieved 91.7%~96.0% in accuracy and 0.952~0.972 in area under the curve using simplified features, including FP1, FP2, F3, O2, and T3 electrodes. Moreover, the combination of greedy stepwise selection and k-nearest neighbor classifier performed better than previously published results. This suggested that fewer EEG channels may be good candidates for usage in portable systems for mild depression detection. As for clinical implementation, Gómez et al. proposed novel findings for assessing changes in EEG activity by a Snoezelen® therapy in “Characterization of EEG patterns in brain-injured subjects and controls after a Snoezelen® intervention [2].” EEG activity was recorded preceding and following a Snoezelen® session in three groups, 18 subjects with cerebral palsy (CP), 18 subjects with traumatic brain-injury (TBI), and 18 controls. The signals were analyzed by means of spectral and nonlinear measures. Experiment results showed decreased values in EEG activity as a consequence of the therapy, and main changes between pre-stimulation and post-stimulation conditions were observed in occipital and parietal brain areas. Additionally, these changes were more widespread in controls than in brain-injured groups due to cognitive deficits in CP and TBI subjects. The findings supported the notion that Snoezelen® therapy induces a slowing of oscillatory activity in central nervous system. For healthcare applications, Contreras et al. developed a new clustering methodology to identify blood glucose (BG) dynamics profiles in “Profiling intra-patient type I diabetes behaviors [3].” Different profiles for diabetic patients were identified from a clustering method based on normalized compression distance, and validated with both in silico and in vivo data. First, in silico experiments were evaluated by 10 patients in 3 different scenarios: days without exercise, poor controlled exercise days, and days with well-controlled exercise. Results demonstrated that the proposed method can identify poor and well-controlled days in theoretical scenarios. Second, in vivo experiments were performed using data from 10 patients to assess performance in real scenarios, and meaningful profiles were presented for working days, bank days, and other situations. This indicated that a tool for profiling BG dynamics can be implemented to enhance existing platforms using continuous glucose monitors and continuous subcutaneous insulin infusion. The above editors’ choice articles incorporated data mining techniques in clinical implementations to predict mild depression, characterize EEG patterns, and identify BG dynamics profiles. In the future, researchers will endeavor to develop more effective data mining systems collectively to facilitate healthcare.