Peri-Diagnostic Decision Support Through Cost-Efficient Feature Acquisition at Test-Time

Abstract

Computer-aided diagnosis (CADx) algorithms in medicine provide patient-specific decision support for physicians. These algorithms are usually applied after full acquisition of high-dimensional multimodal examination data, and often assume feature-completeness. This, however, is rarely the case due to examination costs, invasiveness, or a lack of indication. A sub-problem in CADx, which to our knowledge has received very little attention among the CADx community so far, is to guide the physician during the entire peri-diagnostic workflow, including the acquisition stage. We model the following question, asked from a physician's perspective: "Given the evidence collected so far, which examination should I perform next, in order to achieve the most accurate and efficient diagnostic prediction?". In this work, we propose a novel approach which is enticingly simple: use dropout at the input layer, and integrated gradients of the trained network at test-time to attribute feature importance dynamically. We validate and explain the effectiveness of our proposed approach using two public medical and two synthetic datasets. Results show that our proposed approach is more cost-and feature-efficient than prior approaches and achieves a higher overall accuracy. This directly translates to less unnecessary examinations for patients, and a quicker, less costly and more accurate decision support for the physician.

A systematic review on artificial intelligence techniques for detecting thyroid diseases

“… Mahurkar & Gaikwad (2017) exploit artificial neural networks in conjunction with K-Means to normalize raw data for hypo- and hyperthyroidism. In Vivar et al (2020) , a guiding computer-aided diagnosis system, using a neural network with a dropout at the input layer, and integrated gradients of the trained network at test-time to attribute feature importance dynamically, is proposed. The technique is applied also to the UCI thyroid dataset to detect both hypo- and hyperthyroidism.…”

“…Looking at the public datasets, we observed that the most used dataset is the UCI one ( https://archive.ics.uci.edu/ml/datasets/Thyroid+Disease ), exploited 27 times ( Duggal & Shukla, 2020 ; Shahid et al, 2019 ; Pan et al, 2016 ; Pavya & Srinivasan, 2017 ; Mahurkar & Gaikwad, 2017 ; Ahmed & Soomrani, 2016 ; Tyagi, Mehra & Saxena, 2018 ; Kumar, 2020 ; Pasha & Mohamed, 2020 ; Shen et al, 2016 ; Bentaiba-Lagrid et al, 2020 ; Raisinghani et al, 2019 ; Vivar et al, 2020 ; Li et al, 2019b ; Ma et al, 2018 ; Kour, Manhas & Sharma, 2020 ; Khan, 2021 ; Priyadharsini & Sasikala, 2022 ; Peya, Chumki & Zaman, 2021 ; Chaubey et al, 2021 ; Hosseinzadeh et al, 2021 ; Juneja, 2022 ; Kishor & Chakraborty, 2021 ; Islam et al, 2022 ; Saktheeswari & Balasubramanian, 2021 ; Chandel et al, 2016 ; Priya & Manavalan, 2018 ). The UCI dataset is characterized by 7,200 instances and 21 categorical and real attributes.…”

“…Some authors highlight the necessity to increase the adopted dataset ( Poudel et al, 2019 ; Yin et al, 2019 ; Moran et al, 2018 ; Kaur, Kumar & Kumar, 2019 ; Ma et al, 2018 ; Qin et al, 2021 ; Shen et al, 2021 ). In other studies, authors identify as a limit the reduced heterogeneity and balancing of the adopted dataset ( Pan et al, 2016 ; Yi et al, 2017 ; Pi et al, 2022 ; Vivar et al, 2020 ). Moreover, in the surveyed datasets, the usage of clinical data is prevalent; thus, also in view of the greater usage of various deep learning techniques, greater development of public image datasets to be used with proper deep convolutional neural networks should be encouraged.…”

Using Base-ml to Learn Classification of Common Vestibular Disorders on DizzyReg Registry Data

“…To this end, we compare several linear, non-linear and neural-network based ML algorithms, along with a novel graph deep learning method that we recently proposed (6,12,13). Following insights from multiple classification experiments for diagnostic decision support in our research over the last few years (4,6,13,14), we also provide a multi-faceted analysis of algorithm outcomes, including an examination of class imbalance, multiple classification metrics, patient feature distributions, and feature importances as rated by the classifiers. To alleviate the implementation burden for multi-algorithm comparison and multivariate evaluation, we provide base-ml as an open-source tool 1 to the vestibular research community, as a starting point for further studies in this direction.…”