An MRI-based diagnostic framework for early diagnosis of dyslexia

Purpose A computer-aided diagnosis (CAD) system for early diagnosis of dyslexia was developed and tested. Dyslexia can severely impair the learning abilities of children, so improved diagnostic methods are needed. Neuropathological studies show abnormal anatomy of the cerebral white matter (CWM) in dyslexic brains. We sought to develop an MRI-based macroscopic neuropathological correlate to the minicolumnopathy of dyslexia that relates to cortical connectivity: the gyral window. The brains of dyslexic patients often exhibit decreased gyrifications, so the thickness of gyral CWM for dyslexic subjects is greater than for normal subjects. We developed an MRI-based method for assessment of gyral CWM thickness with automated recognition of abnormal (e. g., dyslexic) brains. Methods In vivo data was collected from 16 right-handed dyslexic men aged 18-40 years, and a group of 14 controls matched for gender, age, educational level, socioeconomic background, handedness and general intelligence. All the subjects were physically healthy and free of history of neurological diseases and head injury. Images were acquired with the same 1.5TMRI scanner (GE, Milwaukee, WI, USA) with voxel resolution 0.9375 x 0.9375 x 1.5 mm using a T1-weighted imaging sequence protocol. The "ground truth" diagnosis to evaluate the classification accuracy for each patient was given by the clinicians. The accuracy of diagnosis/classification of both the training and test subjects was evaluated using the Chi-square test at the three confidence levels-85, 90 and 95%-in order to examine significant differences in the Levy distances. Results As expected, the 85% confidence level yielded the best results, the system correctly classified 16 out of 16 dyslexic subjects (a 100% accuracy) and 14 out of 14 control subjects (a 100% accuracy). At the 90% confidence level, 16 out of 16 dyslexic subjects were still classified correctly; however, only 13 out of 14 control subjects were correct, bringing the accuracy rate for the control group down to 92.86%. The 95% confidence level obviously gives the smaller accuracy rates for both the groups, namely, 14 out of 16 correct answers for dyslexic subjects (87.5%) and still 13 out of 14 control subjects (92.86%). The classification based on traditional volumetric approach is 7 out of 16 dyslexic subjects (a 43.75% accuracy), and 9 out of 14 control subjects (a 64.29% accuracy) at a 85 confidence interval. These results highlight the advantage of the proposed diagnostic approach. Conclusion We found that 3D texture analysis of MRI brain scans can accurately discriminate dyslexic and normal subjects in this feasibility trial. Our method for white matter segmentation and classification outperforms volumetric descriptions of brain structures and may be influenced less by age effects and segmentation errors. The proposed approach efficiently extracts quantitative features from 3D shapes of brain structures.