Cascade Network Model to Detect Cognitive Impairment using Clock Drawing Test

Abstract

The Clock-Drawing Test (CDT) is commonly used to screen people for assessing cognitive impairment. Diagnoses are based on analyzing the specific features of clock drawing with pen and paper. The manual interpretations and understanding of the features are time-consuming, and test results highly depend on clinical experts' knowledge. Due to the impact of smart devices and advancements in deep learning algorithms, the necessity of a consistent and automatic screening system for cognitive impairment has amplified. This work proposed a simple, fast, low-cost, automated CDT screening technique. Initially, transferred deep convolution neural networks (ResNet152, EfficientNetB4, and DenseNet201) are used as feature extractors. The transfer learning technique makes it possible to experiment with existing models and build models much more quickly. Further, the extracted features are cascaded into a feature fusion layer to improve the quality of learning features, and the obtained feature vector become input for the classifier for classification. The performance of the model is experimentally evaluated and compared with the existing state-of-art models on a real dataset. Obtained results demonstrated that the Cascaded Network Model achieves high performance with an accuracy of 97.76%.