Artificial Intelligence in Magnetic Resonance Imaging Laboratory (AIMRI Lab)

歡迎有興趣之學士班大專生、碩博士班學生、專兼任研究助理加入!


The goal of AIMRI lab is to develop noninvasive functional and diffusion MRI methods and imaging markers for early detection of neural plasticity in activity and connectivity to neurocytoarchitecture, and for evaluation of treatment in neurodegenerative diseases. We also use machine learning and deep learning to predict disease status with multiple MRI features.

(I) We employed multiple MRI features, machine learning, and deep learning to predict suicidal attempt, suicidal ideation, depression, chemo-brain in breast cancer survivors, and the effects of betel-quid chewing.

(II) We established neuroimaging markers in various populations, including individuals with major depressive disorder and suicidal tendencies, breast cancer patients, obese individuals, those with tuberous sclerosis complex (TSC), and Chinese-speaking children with specific learning disabilities.

(III) Using generalized q-sampling imaging (GQI), resting-state fMRI, and graph theoretical analysis, we identified gender differences in the structural connectome of the teenage brain and assessed structural and functional connectivity changes in betel-quid chewers.

(IV) By using manganese-enhanced MRI (MEMRI), we demonstrated the neuroprotective effects of Ceftriaxone in a rat model of Parkinson's disease, preventing neurodegeneration and reduced neurogenesis. Additionally, resting-state fMRI was employed to investigate the effects of taurine in spontaneously hypertensive rats.

(V) We employed advanced diffusion methods to examine brain changes in rats and rabbits infected with Angiostrongylus Cantonensis.

(VI) We utilized GQI to evaluate rabbit brain development and assess brain injury following cerebral hemisphere radiation exposure.

(VII) Furthermore, we mapped rat barrel activation after whisker stimulation using activity-induced Manganese-dependent contrast and investigated plasticity in the forepaw digit barrel subfield of rat brains using BOLD fMRI.

Servers & NAS

3T MRI

Servers & NAS

基於多特徵磁振造影與人工智慧之客觀式自殺預警系統

由長庚大學醫放系與嘉義長庚醫院精神科所組成的「人工智慧精神醫學研究團隊」研發「基於多特徵磁振造影與人工智慧之自殺風險評估系統」,已發表二篇國際期刊論文(Weng, Chen et al. 2020, Chen, Weng et al. 2021)與取得二項中華民國新型專利(2020 & 2021)與一項中華民國發明專利(2022),並榮獲第18屆國家新創獎-臨床新創獎(2021)與第19屆國家新創獎-新創精進獎(2022),簡而言之,就是以磁振造影(magnetic resonance imaging, MRI)配合人工智慧方法分析憂鬱、自殺意念及自殺嘗試的風險。在自殺意念及自殺嘗試部分,特異性都超過90%,這是用來輔助醫師診斷重要的指標,也是自殺研究領域追求的終極目標。在醫學上,關於生命的問題,醫界莫不積極開發各種更進階的診斷工具來增加診斷準確性,然而在自殺這個生命交關的問題上,全世界每年有一百萬人死於自殺,至今卻仍然只能仰賴醫療人員的問診而已。我們的研究發現自殺的人腦部神經細胞外層的髓鞘其實已經有多處損傷,用機器學習與深度學習的方式根據這些髓鞘的損傷將其分類,便可大幅度提高對自殺風險評估的準確性。所以下次有人想自殺,就不要教他想開一點,教他想想有多少人愛他,世界有多可貴,如何更勇敢。他的神經都受損了,這些話語就像遠水救不了近火,就讓他能在安全的環境下,可以透過陪伴支持與治療,慢慢的將神經修復,恢復到擁有可以接受這些話語背後意涵的能力與心情。

一個世界衛生組織跨17個國家的國際研究顯示,自殺意念的終生盛行率為9.2%,另一研究顯示一年的盛行率為2% (Nock, Borges et al. 2008, Borges, Nock et al. 2010)。而台灣的全國性調查顯示,一周內有2.8%的人有自殺意念,一年內有5.5%,而終生有18.5%的人有自殺意念(Lee, Lee et al. 2010),大約是國際平均的二倍。自殺意念者有1/3會真正去做自殺嘗試,而一年內會去嘗試的有1/5 (Borges, Nock et al. 2010, Nock, Green et al. 2013),自殺意念評估是自殺防治的重要步驟。而精神疾病,尤其是憂鬱症是自殺最強的危險因子,自殺嘗試者有90%有精神疾病,95%自殺死亡者有精神疾病(Litman 1989, Mościcki 2001)。

自殺死亡的人80%死前一年內看過其他科醫師,只有30%看過精神科醫師(Walby, Myhre et al. 2018, Stene-Larsen and Reneflot 2019),有59%的憂鬱症自殺者會和精神科醫師表達自殺意圖,有19%的憂鬱症自殺者會和非精神科醫師表達(Isometsa, Henriksson et al. 1994)。另外9%的自殺死亡發生在精神科病房出院後的一天內(Pirkis and Burgess 1998),可見專業精神科醫師得評估都無法保證其正確性。目前臨床上自殺風險的評估完全仰賴不同程度的專業醫療人員臨床會談診斷及其他量表輔助的主觀判定,及患者本身的意願,因此是一種評估準確度不穩定的過程。例如2015年外電曾報導一件奇特且極其罕見的謀殺兼自殺案件。德國廉價航空日耳曼之翼的27歲副機師魯比茲(Andreas Lubitz)疑似刻意讓飛機在法國南部阿爾卑斯山區撞毀,機上150人全數罹難,引發外界議論魯比茲是不是謀殺犯。德國檢方宣布魯比茲確實曾有憂鬱自殺傾向,但在最後的治療階段,已没有任何跡象顯示他有傷害自己或他人的傾向。若當時能有一套客觀式自殺預警系統,也許就能避免遺憾發生。開發協助臨床人員的客觀評估工具,如同測量血壓能夠有血壓計一般,成為此領域長期亟待突破的瓶頸。

我們的發明利用機器學習與深度學習對不同嚴重程度之憂鬱症患者做區分,分類採用的特徵為擴散磁振造影影像,篩選出具有自殺企圖、自殺意念與憂鬱之患者,施予適當的醫療照護,達到防範自殺之目的,此發明的特點為我們使用通用擴散波數取樣磁振造影(generalized q-sampling imaging, GQI)與大腦聯結體(brain connectome)分析等技術,能有效偵測自殺企圖、自殺意念與憂鬱之患者腦部神經髓鞘的細微變化,然而該變化在影像呈現上對臨床醫師還是難以直接判讀,因此我們採用機器學習邏輯回歸(logistic regression, LR)與極限梯度提升(extreme gradient boosting, XGB)預測自殺意念,準確率達85% (Weng, Chen et al. 2020),與深度學習DenseNet進行自殺企圖大腦磁振影像分類,準確率達93.7% (Chen, Weng et al. 2021),最終用途是成為臨床篩選篩自殺企圖、自殺意念與憂鬱患者之輔助標準。

衛生署國民健康局以台灣人憂鬱症量表做二萬多人社區人口的調查,可發現15 歲以上民眾8.9%有中度以上憂鬱,5.2%有重度憂鬱。年齡65歲以上8.4%有重度憂鬱,其次15-17歲6.8%有重度憂鬱。台灣的全國性調查顯示,一周內有2.8%的人有自殺意念,一年內有5.5%,而終生有18.5%的人有自殺意念(Lee, Lee et al. 2010),估計憂鬱人口逾百萬,可見市場潛力之大。我們採用腦磁振造影技術為患者進行掃描,獲取初始影像資料後進行影像處理分析,處理完成後,接著以人工智慧進行預測,預測該患者是否有自殺企圖、自殺意念與憂鬱症,最終預測的結果提供給臨床醫師作為輔助診斷。

我們的發明已經取得二項中華民國新型專利(2020 & 2021)與一項中華民國發明專利(2022),將來可將影像處理、機器學習與深度學習模組等軟體商品化或建構成為一雲端平台,並與醫院合作,醫院上傳有效之憂鬱症患者磁振造影影像,資訊平台即可加以分析與預測自殺企圖、自殺意念、憂鬱症等,並可以提供預測結果與治療建議。

參考文獻:

Borges, G., M. K. Nock, J. M. Haro Abad, I. Hwang, N. A. Sampson, J. Alonso, L. H. Andrade, M. C. Angermeyer, A. Beautrais, E. Bromet, R. Bruffaerts, G. de Girolamo, S. Florescu, O. Gureje, C. Hu, E. G. Karam, V. Kovess-Masfety, S. Lee, D. Levinson, M. E. Medina-Mora, J. Ormel, J. Posada-Villa, R. Sagar, T. Tomov, H. Uda, D. R. Williams and R. C. Kessler (2010). "Twelve-month prevalence of and risk factors for suicide attempts in the World Health Organization World Mental Health Surveys." J Clin Psychiatry 71(12): 1617-1628.

Chen, V. C., F. T. Wong, Y. H. Tsai, M. T. Cheok, Y. E. Chang, R. S. McIntyre and J. C. Weng (2021). "Convolutional Neural Network-Based Deep Learning Model for Predicting Differential Suicidality in Depressive Patients Using Brain Generalized q-Sampling Imaging." J Clin Psychiatry 82(2).

Isometsa, E. T., M. M. Henriksson, H. M. Aro and J. K. Lonnqvist (1994). "Suicide in bipolar disorder in Finland." Am J Psychiatry 151(7): 1020-1024.

Lee, J. I., M. B. Lee, S. C. Liao, C. M. Chang, S. C. Sung, H. C. Chiang and C. W. Tai (2010). "Prevalence of suicidal ideation and associated risk factors in the general population." J Formos Med Assoc 109(2): 138-147.

Litman, R. E. (1989). "500 psychological autopsies." J Forensic Sci 34(3): 638-646.

Mościcki, E. K. (2001). "Epidemiology of completed and attempted suicide: toward a framework for prevention." Clinical Neuroscience Research 1(5): 310-323.

Nock, M. K., G. Borges, E. J. Bromet, J. Alonso, M. Angermeyer, A. Beautrais, R. Bruffaerts, W. T. Chiu, G. de Girolamo, S. Gluzman, R. de Graaf, O. Gureje, J. M. Haro, Y. Huang, E. Karam, R. C. Kessler, J. P. Lepine, D. Levinson, M. E. Medina-Mora, Y. Ono, J. Posada-Villa and D. Williams (2008). "Cross-national prevalence and risk factors for suicidal ideation, plans and attempts." Br J Psychiatry 192(2): 98-105.

Nock, M. K., J. G. Green, I. Hwang, K. A. McLaughlin, N. A. Sampson, A. M. Zaslavsky and R. C. Kessler (2013). "Prevalence, correlates, and treatment of lifetime suicidal behavior among adolescents: results from the National Comorbidity Survey Replication Adolescent Supplement." JAMA Psychiatry 70(3): 300-310.

Pirkis, J. and P. Burgess (1998). "Suicide and recency of health care contacts. A systematic review." Br J Psychiatry173: 462-474.

Stene-Larsen, K. and A. Reneflot (2019). "Contact with primary and mental health care prior to suicide: A systematic review of the literature from 2000 to 2017." Scand J Public Health 47(1): 9-17.

Walby, F. A., M. O. Myhre and A. T. Kildahl (2018). "Contact With Mental Health Services Prior to Suicide: A Systematic Review and Meta-Analysis." Psychiatr Serv 69(7): 751-759.

Weng, J. C., T. Y. Lin, Y. H. Tsai, M. T. Cheok, Y. E. Chang and V. C. Chen (2020). "An Autoencoder and Machine Learning Model to Predict Suicidal Ideation with Brain Structural Imaging." J Clin Med 9(3).

MRI & GQI

Tractography

Deep Learning

JC版權所有© 2009-2023 JC All Rights Reserved.