Deep transfer learning for alzheimer neurological disorder detection

Alzheimer's disease is becoming common in the world with the time. It is an irreversible and progressive brain disorder that slowly destroys the memory and thinking skills and, eventually, the ability to perform the simplest tasks. It becomes severe before the noticeable symptoms appear and causes brain disorder which cannot be cured by any medicines and therapies, however its progression can be slow down through early diagnosis. In this paper, we employed different CNN based transfer learning methods for Alzheimer disease classification. We have applied different parameters, and achieved remarkable accuracy on benchmark ADNI dataset. We have tested 13 differnt flavours of different pre-trained CNN models using a fine-tuned approach of transfer learning across two different domain on ADNI dataset (94 AD, 138 MCI and 146 NC). Comparatively, DenseNet showed better performance by achieving a maximal average accuracy of % 99.05. Significant improvement in accuracy has been observed as compared to previously reported works in terms of specificity, sensitivity and accuracy. The source code of propose framework is publicly available.

Automated summary

Alzheimer's disease is a common irreversible brain disorder that affects memory and thinking skills. The study used CNN transfer learning methods to classify Alzheimer's disease and achieved remarkable accuracy on the ADNI dataset, with DenseNet performing the best. The proposed framework's source code is publicly available.