Alzheimer’s Disease Early Detection
Using a Low Cost Three-Dimensional
Densenet-121 Architecture
Braulio Solano-Rojas
(B) , Ricardo Villal´ on-Fonseca
(B) ,
and Gabriela Mar´ ın-Ravent´ os
(B)
CITIC - ECCI, Universidad de Costa Rica, San Jos´ e, Costa Rica
{braulio.solano,ricardo.villalon,gabriela.marin}@ucr.ac.cr
Abstract. The objective of this work is to detect Alzheimer’s disease
using Magnetic Resonance Imaging. For this, we use a three-dimensional
densenet-121 architecture. With the use of only freely available tools, we
obtain good results: a deep neural network showing metrics of 87% accuracy, 87% sensitivity (micro-average), 88% specificity (micro-average),
and 92% AUROC (micro-average) for the task of classifying five different classes (disease stages). The use of tools available for free means that
this work can be replicated in developing countries.
Keywords: Alzheimer · Deep learning · MRI · Computer-aided
detection · Computer-aided diagnosis
1 Introduction
Alzheimer’s Disease (AD) is the most common form of dementia among older
adults [17]. It is a neurodegenerative disease without a cure. Its early detection
is crucial because it allows those people who are going to be affected to prepare
for future changes [17]. For example, some medications delay the disease. Also,
their relatives can prepare and train for the care that will be necessary [17].
Early detection is not easy. One of the difficulties is the performance of people
working at the clinic. People making a diagnosis are affected by several factors
such as fatigue, stress, distractions, and inherent cognitive biases to specific
conditions of the disease. When radiologists see a medical image, such as a
magnetic resonance imaging (MRI), biased reasoning about the conditions of
the disease will result in the loss of the opportunity to detect it. Graber et al. [7]
found that about 74% of diagnostic errors are attributed to cognitive factors. Lee
et al. [14] state that approximately 75% of all medical errors made were due to
diagnostic errors by radiologists. A high workload, stress, fatigue, cognitive bias,
and an inadequate system are part of the causal factors. Medical errors contrast
with the fact that recently artificial intelligence, in particular, deep artificial
Supported by CITIC and ECCI, Universidad de Costa Rica.
c
The Author(s) 2020
M. Jmaiel et al. (Eds.): ICOST 2020, LNCS 12157, pp. 3–15, 2020.
https://doi.org/10.1007/978-3-030-51517-1_1
Using a Low Cost Three-Dimensional
Densenet-121 Architecture
Braulio Solano-Rojas
(B) , Ricardo Villal´ on-Fonseca
(B) ,
and Gabriela Mar´ ın-Ravent´ os
(B)
CITIC - ECCI, Universidad de Costa Rica, San Jos´ e, Costa Rica
{braulio.solano,ricardo.villalon,gabriela.marin}@ucr.ac.cr
Abstract. The objective of this work is to detect Alzheimer’s disease
using Magnetic Resonance Imaging. For this, we use a three-dimensional
densenet-121 architecture. With the use of only freely available tools, we
obtain good results: a deep neural network showing metrics of 87% accuracy, 87% sensitivity (micro-average), 88% specificity (micro-average),
and 92% AUROC (micro-average) for the task of classifying five different classes (disease stages). The use of tools available for free means that
this work can be replicated in developing countries.
Keywords: Alzheimer · Deep learning · MRI · Computer-aided
detection · Computer-aided diagnosis
1 Introduction
Alzheimer’s Disease (AD) is the most common form of dementia among older
adults [17]. It is a neurodegenerative disease without a cure. Its early detection
is crucial because it allows those people who are going to be affected to prepare
for future changes [17]. For example, some medications delay the disease. Also,
their relatives can prepare and train for the care that will be necessary [17].
Early detection is not easy. One of the difficulties is the performance of people
working at the clinic. People making a diagnosis are affected by several factors
such as fatigue, stress, distractions, and inherent cognitive biases to specific
conditions of the disease. When radiologists see a medical image, such as a
magnetic resonance imaging (MRI), biased reasoning about the conditions of
the disease will result in the loss of the opportunity to detect it. Graber et al. [7]
found that about 74% of diagnostic errors are attributed to cognitive factors. Lee
et al. [14] state that approximately 75% of all medical errors made were due to
diagnostic errors by radiologists. A high workload, stress, fatigue, cognitive bias,
and an inadequate system are part of the causal factors. Medical errors contrast
with the fact that recently artificial intelligence, in particular, deep artificial
Supported by CITIC and ECCI, Universidad de Costa Rica.
c
The Author(s) 2020
M. Jmaiel et al. (Eds.): ICOST 2020, LNCS 12157, pp. 3–15, 2020.
https://doi.org/10.1007/978-3-030-51517-1_1