专家与新手放射科医生医学图像视觉搜索差异研究
发布时间：2018-07-25 14:11
【摘要】：现代医学影像技术通过非侵入式的检测方法,提供了关于人体组织、器官的结构和功能图像,已经在医疗诊断中发挥着越来越重要的作用。据统计,单就胸部及乳腺X光片检查而言,每年全球进行约10亿人次检查,每秒平均有120张医学图像产生。影像的迅速发展使得临床医生对病人的病情诊断更加精确,效率更高。但已有研究表明,放射学检查中,大约有20-30%的漏报率。在所有影像检查中,胸部X光片检查应用最广,占所有放射影像检查的40%以上。然而放射科医生在胸部X光片诊断时的漏报率更高(26%以上)。漏报危害到患者的生命安全,引起医疗纠纷,甚至引发重大医疗事故。众多研究表明,专家在医学图像视觉搜索的准确性和反应速度均优于新手。探明专家优异表现背后的认知加工机制有助于提高医疗诊断绩效,为放射学领域提供有效的专家成长模式。本研究将场景知觉、视觉搜索等心理学领域相关的理论和方法应用于医学图像视觉搜索的研究,采用胸部X光片肺结节检测任务,应用专家——新手范式,从视觉复杂度、任务因素和线索提示三个方面探讨专家和新手放射科医生进行医学图像视觉搜索的差异。本论文包括三项研究,共8个实验,均采用视觉搜索范式。研究一通过改变材料的视觉复杂度,探索专家与新手放射科医生进行医学图像视觉搜索的差异,包括3个实验。实验1探讨专家与新手在不同显著结节医学图像上的视觉搜索差异。15名专家和15名新手参加实验,为2(经验:专家、新手)×3(结节显著性:高、中、低)的混合设计。结果发现,专家不仅漏报率更低,且扫描模式更简洁高效。在难度高的实验材料上,专家表现出更大的优势。在漏报类型上,专家倾向于决策错误,新手倾向于扫描错误。同时研究表明,经验和结节显著性共同影响着观察者的搜索绩效以及注视的分配。进一步分析发现,二者在扫描阶段对图像的视觉注视是共同起作用的,而在确认阶段对视觉注视独立起作用。研究还表明经验仅在扫描阶段起促进作用。实验2探讨结节大小和结节位置对专家与新手进行医学图像视觉搜索的影响。15名专家和20名新手参加实验,为2(经验:专家、新手)×2(结节大小:大、小)×2(结节位置:隐蔽位置、非隐蔽位置)的混合设计。结果发现,专家在搜索绩效及加工效率方面均优于新手,且专家的注视偏好位置与重要的解剖学区域有很好的相关。研究还表明,经验和结节大小、结节位置共同影响着观察者的搜索绩效及注视的分配。进一步分析发现,经验和结节大小在扫描阶段对图像的视觉注视是共同起作用的,而在确认阶段二者对视觉注视是独立起作用的。同时研究还表明经验仅在扫描阶段起促进作用。实验3探讨结节大小和结节密度对专家与新手进行医学图像视觉搜索的影响。15名专家和15名新手参加实验,为2(经验:专家、新手)×2(结节大小:大、小)×2(结节密度:高密度、低密度)的混合设计。结果发现,专家在搜索绩效及加工效率方面均优于新手。研究还表明,经验和结节密度共同影响着观察者的搜索绩效以及注视的分配。进一步分析发现,经验和结节密度在扫描阶段对图像的视觉注视是共同起作用的,而在确认阶段二者对视觉注视是独立起作用的。同时研究还表明经验仅在扫描阶段起促进作用。研究二通过操纵不同的时间压力和事件率,探索任务因素对专家与新手放射科医生进行医学图像视觉搜索的影响,揭示漏报错误产生的机制。包括2个实验。实验4探讨专家与新手在不同时间压力条件下的视觉搜索差异。13名专家和13名新手参加实验,为2(经验:专家、新手)×8(时间压力:1-8)×2(结节大小:大、小)的混合设计。研究结果表明,时间压力显著影响专家与新手的漏报错误。8种时间压力条件下,专家的敏感性均好于新手。专家与新手在各时间压力条件下漏报错误的趋势略有不同:大结节条件下,专家与新手均表现出在10s后的漏报错误趋于稳定,小结节条件下,专家在25s后的漏报错误趋于稳定,而新手在20s后的漏报错误趋于稳定。专家与新手均随着时间压力的增加,判断变得更加保守。实验5探讨专家与新手在不同事件率下的视觉搜索差异。15名专家和15名新手参加实验,为2(经验:专家、新手)×2(事件率:50%、10%)×2(结节大小:大、小)的混合设计。研究结果表明新手在大、小结节上的漏报错误均出现低事件率效应,专家在小结节上的漏报错误出现低事件率效应,但受低事件率的影响程度远远小于新手。研究结果还表明,低事件率条件下,扫描错误可能是导致专家与新手漏报错误的主要原因。研究三通过操纵线索提示的不同类型,探讨线索提示对专家与新手放射科医生进行医学图像视觉搜索的影响。包括3个实验。实验6探讨有、无线索提示对专家与新手视觉搜索的影响。15名专家和20名新手参加实验,为2(经验:专家、新手)×2(线索提示类别:有、无)×2(结节大小:大、小)×2(结节位置:隐蔽位置、非隐蔽位置)的混合设计。结果表明,专家在有线索提示下的隐蔽位置小结节上的漏报率低于无线索提示,而新手在有线索提示下的四种类型结节上的漏报率均低于无线索提示。专家受线索提示的影响程度小于新手。有线索提示条件下,专家与新手搜索目标时的注视次数和总注视时间均增多。回视次数说明新手花更多的认知资源对提示语进行加工。实验7探讨线索提示的权威水平对专家与新手视觉搜索的影响。15名专家和19名新手参加实验,为2(经验:专家、新手)×3(线索提示的权威水平:高、低、无)的混合设计。结果表明,新手在高权威线索提示、低权威线索提示下的漏报率均低于无线索提示。专家在高权威线索提示下的漏报率低于无线索提示,但专家受高权威线索提示的影响程度小于新手。高权威线索提示增加了专家的注视次数和总注视时间;新手在高、低权威线索提示下的注视次数和总注视时间显著增多。实验8探讨线索提示的一致性对专家与新手视觉搜索的影响。16名专家和20名新手参加实验,为2(经验:专家、新手)×3(线索提示的一致性:一致、不一致、无)的混合设计。专家在线索提示一致下的漏报率边缘高于线索提示不一致,但二者与无线索提示下的漏报率差异均不显著。在线索提示一致时,专家与新手的漏报率无差异;在线索提示不一致时,专家的漏报率低于新手。新手在线索提示一致时的漏报率低,线索提示不一致时的高。在本论文研究条件下,得出以下结论:(1)医学图像视觉搜索存在专长的知觉编码优势效应,专家不仅漏报率更低、而且搜索目标更快,扫描模式简洁高效,知觉广度更大,可以选择性加工图像信息。(2)专家的优势表现在难度较高的医学图像的视觉搜索。(3)医学图像对目标物体的识别是通过自上而下和自下而上的两条加工通路的整合而实现的。但是知识经验和目标物体特征的交互影响在视觉搜索的不同阶段存在着差异,体现在扫描阶段二者对图像的视觉注视是共同起作用的,而在确认阶段二者对视觉注视是独立起作用的。知识经验仅在扫描阶段起促进作用。(4)专家与新手的漏报错误都受到时间压力的影响,但二者在各时间压力条件下漏报错误的趋势不同。医学图像视觉搜索的时间至少要25s。(5)医学图像视觉搜索呈现低事件率效应特点,新手在大、小结节上的漏报错误均出现低事件率效应,专家在小结节上的漏报错误出现低事件率效应,但专家受低事件率的影响程度小于新手。低事件率条件下,扫描错误可能是导致专家与新手漏报错误的主要原因。(6)放射科医生做决策时,其知识经验与他人的影响共同起作用。他人影响对新手来说作用更大。(7)经验影响了医生的判断倾向,在低显著结节条件下,专家诊断漏报倾向于决策错误,新手倾向于扫描错误。(8)医学图像视觉搜索漏报错误的影响因素研究发现,高视觉复杂度、高时间压力、低事件率、错误的线索会显著增加医学图像视觉搜索中的漏报错误。
[Abstract]:Modern medical imaging technology has been playing a more and more important role in medical diagnosis through non-invasive testing, which has been playing an increasingly important role in medical diagnosis. According to statistics, about 1 billion times a year, there are about 120 medical images per second on the basis of the single chest and breast X ray examination. Emergence. The rapid development of the image makes the clinician more accurate and more efficient in diagnosing the patient's condition. However, there has been a study showing that there is a 20-30% leakage rate in the radiological examination. In all imaging examinations, the chest X ray examination is the most widely used, accounting for more than 40% of all radiographic examinations. However, the radiologist is in the chest X ray. The leakage rate of the diagnosis is higher (more than 26%). The missing report endangers the patient's life safety, causes medical disputes and even causes major medical accidents. Many studies show that the accuracy and reaction speed of the visual search of medical images are better than those of the novice. The cognitive processing mechanism behind the excellent performance of the experts is helpful to improve the medical diagnosis. Performance, providing an effective expert growth model for the field of radiology. This study applies the theories and methods related to the field of psychology, such as scene perception and visual search, to the research of medical image visual search, using the chest X light slice lung nodule detection task, and applying the expert - Novice paradigm from visual complexity, task factors and clue hints. The three aspect discusses the differences between the experts and the novice radiologists for the visual search of medical images. This paper includes three studies, with a total of 8 experiments, all using visual search paradigm. The first is to explore the differences in visual search of medical images by the experts and the novice radiologists, including 3 experiments. Experiment 1 explored the visual search differences between experts and novice in different significant nodular medical images by.15 experts and 15 novice participants, for 2 (experience: expert, novice) x 3 (nodular significance: high, medium, low). The results showed that the experts not only had lower missing rates, and the scanning mode was more concise and efficient. The difficult experimental materials were more difficult. Experts show a greater advantage. In the type of leakage, experts tend to make mistakes in decision-making and novice tends to scan errors. Meanwhile, the study shows that experience and nodular significance have a common influence on the observer's search performance and the distribution of gaze. Further analysis shows that the visual fixation of the images is the common effect of the two in the scanning phase. The study also showed that experience only played a role in the scanning phase. Experiment 2 explored the effects of nodule size and nodule position on visual search for medical images by experts and novice.15 experts and 20 novice participants, for 2 (experience: expert, novice) * 2 (large, small) * 2 (large, small). Nodular position: a mixed design of concealed position and non concealed position. Results show that experts are superior to novice in both search performance and processing efficiency, and the position of expert's fixation preference is well related to important anatomical areas. The distribution of fixations. Further analysis shows that experience and nodule size are common to the visual gaze of the image at the scanning stage, while the two in the confirmation stage are independent of visual fixation. At the same time, the study also shows that experience only plays a role in the scanning phase. Experiment 3 explores the size and density of nodules for experts and new Hand performed visual search for medical images,.15 experts and 15 newcomers participated in the experiment, and a hybrid design for 2 (experience: expert, novice) x 2 (nodular size: large, small) x 2 (nodular density: density, low density). The results showed that experts were superior to the novice in terms of search performance and efficiency. The study also showed that experience and nodular density were common. It affects the performance of the observer and the distribution of the gaze. Further analysis shows that experience and nodule density play a role in the visual gaze of the image at the scanning stage, while the two in the confirmation stage are independent of visual fixation. At the same time, the study also shows that experience only promotes the scan stage. Study two through the operation. Different time pressure and event rate, explore the effect of task factors on the visual search of medical images by experts and novice radiologists, and reveal the mechanism of missing report errors. Including 2 experiments. Experiment 4 discuss the.13 name experts and 13 novice experts and novice experts and novice experts and novice candidates under different time pressure conditions. A mixed design of 2 (experience: expert, novice) x 8 (time pressure: 1-8) x 2 (nodular size: large, small). The results show that the time pressure significantly affects the.8 time pressure of the experts and novice, the sensitivity of the expert is better than that of the novice. Under the condition of large nodules, both experts and novice showed a tendency to be stable after 10s. Under the condition of small nodules, the mistakes of the experts after 25s tended to be stable, and the novice's error after 20s tended to be stable. Both experts and novice were more conservative with the increase of time pressure. Experiment 5 explored the differences between experts and novice. Visual search differences under event rates.15 experts and 15 novice participants were experimentation, for 2 (experience: expert, novice) x 2 (event rate: 50%, 10%) x 2 (nodular size: large, small). The results showed that novice's leakage errors on large and small nodules all had low event rate effects, and the expert's leakage error on small nodules appeared low events. Rate effect, but the effect of low event rate is far less than that of novice. The results also show that scanning errors may be the main cause of the mistake of the experts and novice in low event rate. Study three explore clues to the medical images of experts and novice radiologists through the different types of manipulating clues. The effect of sense search. Including 3 experiments. Experiment 6 explored the effects of no clues on the visual search of experts and novice,.15 experts and 20 novice students participated in the experiment, for 2 (experience: expert, novice) x 2 (cue category: no) x 2 (nodule size: large, small) x 2 (nodule location: concealed position, non concealed position). The rate of leakage on the small nodules on the hidden location under clues was lower than that without clues, and the novice's rate of leakage on the four types of nodules under clues was lower than that without clues. The degree of influence of the clues prompted by the expert was less than that of the novice. The number and the total fixation time were increased. The number of recalling times indicated that the novice spent more cognitive resources on the processing of the hint. Experiment 7 explored the influence of the authority level on the visual search of the experts and the novice,.15 experts and 19 novice participants, for 2 (experience: expert, novice) * 3 (the level of clues: high, low, and no) mixed. The results show that the novice's reporting rate under the hint of low authoritative clues is lower than that without clue. The leakage rate of experts under the hint of high authoritative clues is lower than that without clue, but the influence degree of the expert is less than that of the novice. The total fixation time; the number of fixations and the total fixation time of the novice at high and low authoritative clues. Experiment 8 explored the effect of consistency on the visual search of experts and novice,.16 experts and 20 novice participants participated in the experiment, and 2 (experience: expert, novice) * 3 (consistency, inconsistency, and no) of the 2 (experience: expert, novice) The leakage rate of the expert online cue is higher than that of the clue, but the difference of the leakage rate between the two and the uncue is not significant. The leakage rate of the expert and the novice is not different when the online cable hints are consistent. When the online cable hints are inconsistent, the rate of the leakage of the experts is lower than that of the novice. Under the conditions of this paper, the following conclusions are obtained: (1) the visual search of medical images has the advantage effect of special perceptual coding, the expert not only has lower false rate, but also the search target is faster, the scanning mode is simple and efficient, the perceptual span is greater, and the image information can be processed selectively. (2) experts (2) experts The advantages are visual search for difficult medical images. (3) the recognition of target objects by medical images is achieved through the integration of two processing paths from top to bottom and bottom-up. However, the interaction between knowledge experience and target object features varies in different stages of visual search, which is reflected in the scanning order. The visual fixation of the two segments is common to the visual fixation of the image, while the two parties play an independent role in the visual fixation during the confirmation stage. Knowledge and experience can only play a role in the scanning phase. (4) the errors of the experts and the novice are all influenced by the time pressure, but the trend of the two in each time pressure condition is different. The time of visual search, like visual search, is at least 25s. (5) of medical image visual search, which shows low event rate effect. The novice's leakage error on large and small nodules all have low event rate effect, and the expert's leakage error on small nodules appears low event rate effect, but the degree of low event rate is less than the novice. Under the low event rate condition, Scanning errors may be the main cause of errors that lead to an error between experts and novice. (6) when a radiologist makes a decision, his knowledge and experience work together with the influence of others. The influence of others is more important to the novice. (7) experience affects the doctor's judgment tendency. Under the condition of low significant nodules, the expert's diagnosis is prone to error decision. The novice is inclined to scan the error. (8) research on the influence factors of medical image visual search leakage error, it is found that high visual complexity, high time pressure, low event rate, and wrong clue can significantly increase the error of medical image visual search.
【学位授予单位】：天津师范大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41
，
本文编号：2144066
[Abstract]:Modern medical imaging technology has been playing a more and more important role in medical diagnosis through non-invasive testing, which has been playing an increasingly important role in medical diagnosis. According to statistics, about 1 billion times a year, there are about 120 medical images per second on the basis of the single chest and breast X ray examination. Emergence. The rapid development of the image makes the clinician more accurate and more efficient in diagnosing the patient's condition. However, there has been a study showing that there is a 20-30% leakage rate in the radiological examination. In all imaging examinations, the chest X ray examination is the most widely used, accounting for more than 40% of all radiographic examinations. However, the radiologist is in the chest X ray. The leakage rate of the diagnosis is higher (more than 26%). The missing report endangers the patient's life safety, causes medical disputes and even causes major medical accidents. Many studies show that the accuracy and reaction speed of the visual search of medical images are better than those of the novice. The cognitive processing mechanism behind the excellent performance of the experts is helpful to improve the medical diagnosis. Performance, providing an effective expert growth model for the field of radiology. This study applies the theories and methods related to the field of psychology, such as scene perception and visual search, to the research of medical image visual search, using the chest X light slice lung nodule detection task, and applying the expert - Novice paradigm from visual complexity, task factors and clue hints. The three aspect discusses the differences between the experts and the novice radiologists for the visual search of medical images. This paper includes three studies, with a total of 8 experiments, all using visual search paradigm. The first is to explore the differences in visual search of medical images by the experts and the novice radiologists, including 3 experiments. Experiment 1 explored the visual search differences between experts and novice in different significant nodular medical images by.15 experts and 15 novice participants, for 2 (experience: expert, novice) x 3 (nodular significance: high, medium, low). The results showed that the experts not only had lower missing rates, and the scanning mode was more concise and efficient. The difficult experimental materials were more difficult. Experts show a greater advantage. In the type of leakage, experts tend to make mistakes in decision-making and novice tends to scan errors. Meanwhile, the study shows that experience and nodular significance have a common influence on the observer's search performance and the distribution of gaze. Further analysis shows that the visual fixation of the images is the common effect of the two in the scanning phase. The study also showed that experience only played a role in the scanning phase. Experiment 2 explored the effects of nodule size and nodule position on visual search for medical images by experts and novice.15 experts and 20 novice participants, for 2 (experience: expert, novice) * 2 (large, small) * 2 (large, small). Nodular position: a mixed design of concealed position and non concealed position. Results show that experts are superior to novice in both search performance and processing efficiency, and the position of expert's fixation preference is well related to important anatomical areas. The distribution of fixations. Further analysis shows that experience and nodule size are common to the visual gaze of the image at the scanning stage, while the two in the confirmation stage are independent of visual fixation. At the same time, the study also shows that experience only plays a role in the scanning phase. Experiment 3 explores the size and density of nodules for experts and new Hand performed visual search for medical images,.15 experts and 15 newcomers participated in the experiment, and a hybrid design for 2 (experience: expert, novice) x 2 (nodular size: large, small) x 2 (nodular density: density, low density). The results showed that experts were superior to the novice in terms of search performance and efficiency. The study also showed that experience and nodular density were common. It affects the performance of the observer and the distribution of the gaze. Further analysis shows that experience and nodule density play a role in the visual gaze of the image at the scanning stage, while the two in the confirmation stage are independent of visual fixation. At the same time, the study also shows that experience only promotes the scan stage. Study two through the operation. Different time pressure and event rate, explore the effect of task factors on the visual search of medical images by experts and novice radiologists, and reveal the mechanism of missing report errors. Including 2 experiments. Experiment 4 discuss the.13 name experts and 13 novice experts and novice experts and novice experts and novice candidates under different time pressure conditions. A mixed design of 2 (experience: expert, novice) x 8 (time pressure: 1-8) x 2 (nodular size: large, small). The results show that the time pressure significantly affects the.8 time pressure of the experts and novice, the sensitivity of the expert is better than that of the novice. Under the condition of large nodules, both experts and novice showed a tendency to be stable after 10s. Under the condition of small nodules, the mistakes of the experts after 25s tended to be stable, and the novice's error after 20s tended to be stable. Both experts and novice were more conservative with the increase of time pressure. Experiment 5 explored the differences between experts and novice. Visual search differences under event rates.15 experts and 15 novice participants were experimentation, for 2 (experience: expert, novice) x 2 (event rate: 50%, 10%) x 2 (nodular size: large, small). The results showed that novice's leakage errors on large and small nodules all had low event rate effects, and the expert's leakage error on small nodules appeared low events. Rate effect, but the effect of low event rate is far less than that of novice. The results also show that scanning errors may be the main cause of the mistake of the experts and novice in low event rate. Study three explore clues to the medical images of experts and novice radiologists through the different types of manipulating clues. The effect of sense search. Including 3 experiments. Experiment 6 explored the effects of no clues on the visual search of experts and novice,.15 experts and 20 novice students participated in the experiment, for 2 (experience: expert, novice) x 2 (cue category: no) x 2 (nodule size: large, small) x 2 (nodule location: concealed position, non concealed position). The rate of leakage on the small nodules on the hidden location under clues was lower than that without clues, and the novice's rate of leakage on the four types of nodules under clues was lower than that without clues. The degree of influence of the clues prompted by the expert was less than that of the novice. The number and the total fixation time were increased. The number of recalling times indicated that the novice spent more cognitive resources on the processing of the hint. Experiment 7 explored the influence of the authority level on the visual search of the experts and the novice,.15 experts and 19 novice participants, for 2 (experience: expert, novice) * 3 (the level of clues: high, low, and no) mixed. The results show that the novice's reporting rate under the hint of low authoritative clues is lower than that without clue. The leakage rate of experts under the hint of high authoritative clues is lower than that without clue, but the influence degree of the expert is less than that of the novice. The total fixation time; the number of fixations and the total fixation time of the novice at high and low authoritative clues. Experiment 8 explored the effect of consistency on the visual search of experts and novice,.16 experts and 20 novice participants participated in the experiment, and 2 (experience: expert, novice) * 3 (consistency, inconsistency, and no) of the 2 (experience: expert, novice) The leakage rate of the expert online cue is higher than that of the clue, but the difference of the leakage rate between the two and the uncue is not significant. The leakage rate of the expert and the novice is not different when the online cable hints are consistent. When the online cable hints are inconsistent, the rate of the leakage of the experts is lower than that of the novice. Under the conditions of this paper, the following conclusions are obtained: (1) the visual search of medical images has the advantage effect of special perceptual coding, the expert not only has lower false rate, but also the search target is faster, the scanning mode is simple and efficient, the perceptual span is greater, and the image information can be processed selectively. (2) experts (2) experts The advantages are visual search for difficult medical images. (3) the recognition of target objects by medical images is achieved through the integration of two processing paths from top to bottom and bottom-up. However, the interaction between knowledge experience and target object features varies in different stages of visual search, which is reflected in the scanning order. The visual fixation of the two segments is common to the visual fixation of the image, while the two parties play an independent role in the visual fixation during the confirmation stage. Knowledge and experience can only play a role in the scanning phase. (4) the errors of the experts and the novice are all influenced by the time pressure, but the trend of the two in each time pressure condition is different. The time of visual search, like visual search, is at least 25s. (5) of medical image visual search, which shows low event rate effect. The novice's leakage error on large and small nodules all have low event rate effect, and the expert's leakage error on small nodules appears low event rate effect, but the degree of low event rate is less than the novice. Under the low event rate condition, Scanning errors may be the main cause of errors that lead to an error between experts and novice. (6) when a radiologist makes a decision, his knowledge and experience work together with the influence of others. The influence of others is more important to the novice. (7) experience affects the doctor's judgment tendency. Under the condition of low significant nodules, the expert's diagnosis is prone to error decision. The novice is inclined to scan the error. (8) research on the influence factors of medical image visual search leakage error, it is found that high visual complexity, high time pressure, low event rate, and wrong clue can significantly increase the error of medical image visual search.
【学位授予单位】：天津师范大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP391.41
，
本文编号：2144066
本文链接：https://www.wllwen.com/guanlilunwen/jixiaoguanli/2144066.html
