與遺傳基因有關的聽力結果和類突發性聽力損失的預測研究

Abstract

聽力損失是感官障礙的一種，通常是環境或遺傳因素引起。其中，遺傳因素在先天性和早發性聽力損失案例中佔有很大的比例。理解特定基因型的表現有助於我們提早進行針對性治療或投藥。SLC26A4是常見的遺傳性聽力損失基因之一。已有多個研究討論了其他基因型，例如GJB2和MYO15A在聽力表現方面的影響。而在本論文的主要目的是利用機器學習技術來探討SLC26A4，以預測病患未來的聽力狀況，並針對可能發生的急性聽力損失進行預測。我們將使用類急性聽力損失為預測目標來討論波動型聽力損失在未來半年內的發生機率。我們利用病患過去的看診聽力檢查數據以及相關的臨床病徵作為特徵，並將類似急性聽力損失的狀況作為預測標籤，以此來探討機器學習在預測波動型聽力損失發生與否方面的能力。在本論文中使用的資料來自具有SLC26A4基因型的病患，其回診追蹤數據具有不定次數和不定間隔的特點。我們的研究包括資料集的建立、缺失值處理、特徵提取和模型架構的建立。最終，我們將會展示在該資料集上的預測能力。

Hearing loss is a sensory impairment, typically caused by environmental or genetic factors. Genetic factors are a significant contributing factor in cases of congenital and early-onset hearing loss. Understanding the clinical manifestations of specific genotypes helps us to initiate targeted treatments or medication at an early stage. SLC26A4 is one of the common genes associated with hereditary hearing loss. Numerous studies have investigated the impact of other genotypes, such as GJB2 and MYO15A, on hearing performance. The main goal of this paper is to explore SLC26A4 using machine learning techniques to predict the future hearing conditions of patients and to forecast the occurrence of acute hearing loss. We will use pseudo-sudden hearing loss as a predictive target to discuss the probability of fluctuating hearing loss over the next six months. We utilize past audiometric examination data and related clinical symptoms of patients as features, and we treat conditions similar to acute hearing loss as predictive labels, to explore machine learning's ability to predict occurrence of fluctuating hearing loss. The data used in this study is sourced from patients with the SLC26A4 genotype, characterized by irregular follow-up frequencies and intervals. Our research encompasses dataset construction, missing value handling, feature extraction, and model architecture development and optimization. Ultimately, we will demonstrate our predictive capability on this dataset.