Abstract
BACKGROUND: Restoring movement after hemiparesis caused by stroke is an ongoing challenge in the field of rehabilitation. With several therapies in use, there is no definitive prescription that optimally maps parameters of rehabilitation with patient condition. Recovery gets further complicated once patients enter chronic phase. In this paper, we propose a rehabilitation framework based on computational modeling, capable of mapping patient characteristics to parameters of rehabilitation therapy. METHOD: To build such a system, we used a simple convolutional neural network capable of performing bilateral reaching movements in 3D space using stereovision. The network was designed to have bilateral symmetry to reflect the bilaterality of the cerebral hemispheres with the two halves joined by cross-connections. This network was then modified according to 3 chosen patient characteristics-lesion size, stage of recovery (acute or chronic) and structural integrity of cross-connections (analogous to Corpus Callosum). Similarly, 3 parameters were used to define rehabilitation paradigms-movement complexity (Exploratory vs Stereotypic), hand selection mode (move only affected arm, CIMT vs move both arms, BMT), and extent of plasticity (local vs global). For each stroke condition, performance under each setting of the rehabilitation parameters was measured and results were analyzed to find the corresponding optimal rehabilitation protocol. RESULTS: Upon analysis, we found that regardless of patient characteristics network showed better recovery when high complexity movements were used and no significant difference was found between the two hand selection modes. Contrary to these two parameters, optimal extent of plasticity was influenced by patient characteristics. For acute stroke, global plasticity is preferred only for larger lesions. However, for chronic, plasticity varies with structural integrity of cross-connections. Under high integrity, chronic prefers global plasticity regardless of lesion size, but with low integrity local plasticity is preferred. CONCLUSION: Clinically translating the results obtained, optimal recovery may be observed when paretic arm explores the available workspace irrespective of the hand selection mode adopted. However, the extent of plasticity to be used depends on characteristics of the patient mainly stage of stroke and structural integrity. By using systems as developed in this study and modifying rehabilitation paradigms accordingly it is expected post-stroke recovery can be maximized.