- Email: [email protected]
A computer support system for neurological anatomical diagnosis
Clinical Neurology and Neurosurgery 110 (2008) 973–978
Contents lists available at ScienceDirect
Clinical Neurology and Neurosurgery journal homepage: www.elsevier.com/locate/clineuro
A computer support system for neurological anatomical diagnosis Hisaki Kamo a,∗ , Yosihisa Kiriyama b,1 , Akihiro Mizoe b,1 , Eiko Murase c,2 , Seiichiro Okajima d,3 , Ichiro Akiguchi e,4 , Yasusuke Hirasawa f,5 , Patrick L. McGeer g,6 a
Department of Neurology, Uzumasa Hospital, 30 Katabiranotujimachi, Uzumasa, Ukyo-Ku, Kyoto City 616-8151, Japan System Laboratory Murata Co., Ltd., 18-8 Kinoshita-cho, Otsu City, Shiga Prefecture 520-0812, Japan c Department of Neurology, Utano National Hospital Organization, 8 Ondoyama-cho, Narutaki, Ukyo-ku, Kyoto City 616-8255, Japan d Department of Orthopedics, Shakaihoken Kobe Central Hospital, 1-1, 2 choume, Souzan-chou, Kita-ku, Koube City, Hyougo Prefecture, Japan e Department of Neurology, Koseikai Takeda Hospital, 841-5, Higashishionokouji-cho, Higashiiru, Shionokoji-dori Nishinotouin, Shimokyou-ku, Kyoto City 600-8558, Japan f Graduate School, Meiji University of Integrative Medicine, Hiyoshi-cho, Funai-gun, Kyoto City 629-0392, Japan g Kinsmen Laboratory of Neurological Research, University of British Columbia, 2255 Wesbrook Mall, Vancouver, BC, Canada V6T 1Z3 b
a r t i c l e
i n f o
Article history: Received 25 January 2008 Received in revised form 10 April 2008 Accepted 27 May 2008 Keywords: Neurological diagnosis Neuroanatomy Anatomical diagnosis Computer Central nervous system Peripheral nervous system Automatic depicting system
a b s t r a c t Objectives: Accurate neurological diagnoses are often difficult to make due to the complexity of the neuroanatomy involved. This study was performed to evaluate the usefulness of a computer system with easily retrievable anatomical information as a support for arriving at more accurate anatomic diagnoses. Patients and methods: Anatomical information from an initial physical examination was programmed into a computer with stored neuroanatomical charts of the brain, spinal cord and peripheral nerves. The information generated a graphic display of possible lesions with suggestions for further examination. These suggestions were then followed and further data entered. This data entry generated a new graphic display with reduced lesion possibilities. Iterations were then followed to narrow the possibilities for diagnosis further, until a final anatomical diagnosis was reached. This method was applied to three hypothetical examples and a number of clinical cases. Here we report three clinical cases in which this method was particularly useful in making a diagnosis. Results: Using computer iterations, the system was able to pinpoint one or more presumptive causative lesions in the CNS or PNS based on known neuronal pathways or nuclei. Conclusion: The results indicate that suitably used, computer memory, by virtue of its large capacity, accuracy and fast recall, can supplement human memory in reaching accurate anatomical diagnoses of neurological lesions. © 2008 Elsevier B.V. All rights reserved.
1. Introduction Accurate diagnoses of neurological diseases can be difficult to make due to the complexity of human neuroanatomy. Furthermore, the problem can be compounded when multiple causative neuronal pathways and nuclei are involved. While magnetic resonance imaging (MRI) and computed tomography (CT) images are frequently relied upon for accurate diagnostic
∗ Corresponding author. Tel.: +81 75 871 0505. E-mail address: [email protected] (H. Kamo). 1 Tel.: +81 77 521 7020. 2 Tel.: +81 75 461 5121. 3 Tel.: +81 78 594 2211. 4 Tel.: +81 75 361 1351. 5 Tel.: +81 771 72 1181. 6 Tel.: +1 604 822 7377. 0303-8467/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.clineuro.2008.05.022
purposes, they may reveal multiple abnormalities making pinpointing of the actual causative lesion difficult. Pattern recognition of neurological signs and symptoms is another approach to accurate diagnosis. Although this approach might seem straightforward, it actually requires a very accurate recall of many specific anatomical details related to numerous disease entities. There may also be significant difficulties when dealing with unusual diseases or unusual manifestations of common diseases [1]. We have studied the possibility of using computer assistance to map the neuronal pathways involved in a patient’s neurological symptoms as a means of accurately localizing the lesions and thus contributing to accurate diagnoses [1–3]. A computer software program was generated in which neuroanatomical charts of the brain, spinal cord and peripheral nerves were stored. Specific charts could then be displayed following input prompting from the clinician according to the physical examination of the patient.
974
H. Kamo et al. / Clinical Neurology and Neurosurgery 110 (2008) 973–978
The principal purpose of this study was to evaluate the ability of this computer program to assist the clinician to pinpoint lesions by displaying information consistent with the patient’s neurological symptoms. Several trials have been performed mostly on pattern recognition diagnoses in restricted areas of the PNS [4–7] and the CNS [8–12] such as the brachial plexus or the brain stem. These were mainly done in the 1970s and 1980s. However, a comprehensive system has not yet been developed for common usage. Advances in computer technology have now made it practical to develop such a system and a prototype version is presented here.
2. Methods A computer program was developed where neuroanatomical charts of the brain, spinal cord [13] and peripheral nerves [14,15] could be stored. Microsoft Visual C++ .NET 2003 and Access 2000 were used to produce all drawings. A depicting system was then programmed to display appropriate neuronal pathways following user input of the patient’s signs and symptoms. For example, in the peripheral motor system, the innervation of each muscle was stored from the nerve root to its distal end. The system was then programmed to display possible lesions corresponding to weaknesses or abnormal neurological findings from EMG test results of each muscle. For the sensory system, all areas of the body surface were subdivided into minimal cutaneous field units corresponding to the sensory spinal roots (dermatome) and peripheral nerves. The software was programmed to depict the responsible pathways, using colored shading from light blue to red to indicate the degree of overlap in cases with more than one symptom. In addition to the depicting systems described above, the software was written to eliminate intact parts of the motor system thus enabling the narrowing down of causative lesions [3,5]. For the central nervous system, the software was programmed to trace the distance along causative neuronal pathways or nuclei on 25 axial sections of the brain and spinal cord, and to indicate the perimeter area excluding all unrelated symptoms [1].
Fig. 1. Computer display following operator input of the symptoms of the Millard–Gubler syndrome. It displays in brown the minimal polygonal area in the lower pons that includes two corticospinal pathways, the facial nerve and the abducens nerve. (a) Facial nerve (b) abducens nerve (c,d) corticospinal tract to upper (c) and lower extremity (d).
3.1.2. Example 2, weakness involving the peripheral motor system In the peripheral motor system, neuronal pathways corresponding to any muscle weakness or EMG abnormality can be displayed. Overlapping pathways are demonstrated by differences in color corresponding to the extent of overlap. Neuronal pathways to normal muscles are shown as broken lines. For example, in a hypothetical case with weakness at the thumb interphalangeal (IP) joint and distal interphalangeal (DIP) joint of forefinger, but without involvement of the metacarpophalangeal (MCP) joint of the thumb (the so-called pinch sign in the anterior interosseous nerve syndrome), the system displays, as the causative lesion, that part of the arm between the start of interosseous nerve and the point of branching to the long flexor muscle of thumb. The unaffected nerve is shown by the broken line (Fig. 3).
3. Results 3.1. Hypothetical examples 3.1.1. Example 1, motor paralysis of CNS origin In a hypothetical examination, clinical symptoms of upper and lower right sided extremity paralysis were observed. The data were entered, and the system responded immediately by displaying the corresponding neuronal pathways on 25 sectional diagrams of the brain and spinal cord. Iteration through additional input was then carried out to allow the potential causative lesion to be deduced. For instance, if the case also had left subnuclear facial paralysis and left abducens palsy in addition to the extremity paralysis, these could be entered into the system described above. All the responsible neuronal pathways were then drawn on a single section of the pons. A polygonal area covering the lower and ventral pontine region corresponding to the Millard–Gubler syndrome was then displayed (Fig. 1). In other circumstances, several levels might be involved. Fig. 2, for example, illustrates the results of such a case. Multiple areas for a causative lesion which produces right motor paralysis plus impairment of thermal-pain and deep sensation are displayed.
3.1.3. Example 3, sensory disturbance in the peripheral sensory system For the peripheral sensory system, the computer displays the whole body showing the appropriate neuronal pathways, with graded coloring indicating the degree of overlap of innervation. In a hypothetical case where disturbed sensation from the thumb to the lateral forearm is presented, which is typical of a C6 radiculoneuropathy, three peripheral nerves might be involved. These are the radial nerve, the median nerve and the lateral antebrachial cutaneous nerve. These are displayed in Fig. 4A. Responsible pathways from the distal end of a nerve to its root are drawn by the subdivided method described above, and the root segment (arrow in Fig. 4B) is identified as the site of the causative lesion, following triple iteration. 3.2. Clinical case examples 3.2.1. Case 1, a CNS lesion A 51-year-old, right-handed man experienced sudden dysphagia, dysarthria and vomiting at a company meeting. He complained of numbness of the right hand, clumsiness of the left hand, and mild unsteadiness when walking. He went to see his family doc-
H. Kamo et al. / Clinical Neurology and Neurosurgery 110 (2008) 973–978
975
Fig. 3. An inferred example of the motor system of peripheral nerves with the anterior interosseous nerve syndrome. The intact median nerve is shown as a broken line and can be eliminated. (a) Median nerve. The affected area of the nerve is displayed as a solid line (b) anterior interosseous nerve (c) nerve to muscle of flexor pollicis longus (d) nerve to muscle of flexor digitorum profundus. Fig. 2. A case with multiple inferred areas. See Section 3 for details of symptoms. (A) Middle pons (B) lower pons (C) upper medulla oblongata. (a) Corticospinal tract (b) lateral spinothalamic tract (c) dorsal column – medial lemniscus pathway.
tor, accompanied by his colleagues, but no abnormal findings were detected by CT that day or by MRI several days later. Since no lesion was identified, a psychosomatic disorder or masked depression was suspected. Accordingly, antidepressant and tranquilizer drugs were prescribed. For several weeks there was only slight improvement in the dysphagia, dysarthria and gait disturbance. Facial and right leg numbness began to develop. As a result, and due to his anxiety, he visited our hospital. Neurological examination revealed the following positive findings: sensory disturbance of pain and temperature in the right upper and lower extremities and left face. There was no impairment of touch or deep sensation, but there was loss of the left gag reflex, mild pharyngeal dysarthria and dysdiadochokinesis of the left hand plus a slight gait ataxia. There was no weakness or rigidity of the upper and lower extremities.
Following entry of these findings into the computer, the causative lesion was demonstrated to be in the left lateral medulla oblongata corresponding to the Wallenberg syndrome (Fig. 5A). Routine MRI tests had been unable to detect the lesion. However, thin slice examination revealed a T1-low, T2-high area in the lateral medulla oblongata confirming the lesion identified by the computer software (Fig. 5B). 3.2.2. Case 2, a motor PNS lesion A 36-year-old man developed right elbow pain and was unable to extend all the fingers of his right hand but was still able to extend his wrist (Fig. 6A). He visited a nearby hospital to consult with an orthopedic surgeon. He was diagnosed as having radial nerve palsy, but due to an inability to localize the lesion precisely, he was treated conservatively for about 4 months. Since he experienced little improvement in his loss of motor function, he visited our hospital. In order to determine the level of the radial nerve lesion, the programmed software was used to anatomically simulate the muscle disturbance
976
H. Kamo et al. / Clinical Neurology and Neurosurgery 110 (2008) 973–978
Fig. 4. Computer display of a sensory nerve disturbance. (A) Area of sensory disturbance (red area) in C6 radiculopathy is displayed on the subdivided body surface with spinal dermatome (red line) and peripheral nerve innervation (blue line). Supply from (a) C6 and lateral antebrachial cutaneous nerve (b) C6 and radial nerve (c) C6 and median nerve (d) C7 and median nerve. (B) Corresponding neuronal pathways of the three divisions overlap at the portion close to the C6 root.
Fig. 5. A clinical case of the Wallenberg syndrome. (A) Computer display of the neuronal pathways corresponding to the symptoms of case 1. (a, b) Lateral spinothalamic tract from upper (a) and lower (b) extremities. (c) Spinal trigeminal tract (d) motor nerve from the ambiguous nucleus. (B) Thin slice MRI image of the medulla oblongata in case 1. A high intensity area was observed in the dorsolateral region (arrow) of the T2 image.
anticipated for lesions at various levels, prior to examining the patient. The muscles chosen were as follows: deltoid, triceps brachii, supinator, extensor carpi radialis longus and brevis (ECR), extensor carpi ulnaris (ECU), extensor pollicis longus (EPL), and the extensor digitorum communis (EDC). Manual muscle testing (MMT) of all these candidate muscles revealed that weakness was confined to the last four muscles. Especially at the wrist, the two carpal extension muscles were found to be particularly important according to the programmed software. MMT was therefore done in the radial direction and the ulnar direction respectively. The result was 5/5 in the radial direction but only 1/5 in the ulnar direction. The results were entered into the software program and the causative lesion was diagnosed as occurring in the upper part of the posterior interosseous nerve (Fig. 6B). EMG testing then demonstrated a denervation potential using fibrillation voltage or positive sharp wave in the resting state for
the muscles of ECU, EPL and EDC, but not for brachioradialis, ECR, deltoid, or triceps. Only minimal discharge could be found in the muscles of ECU, and no reinnervation potential was found in EPL and EDC following voluntary contraction. In conclusion, little tendency for recovery was shown in the electromyographic tests. SEP tests demonstrated normal findings and cervical MRI showed only mild disc bulging at C6/7 without canal or intervertebral canal stenosis. The result of sphygmography was normal. Using these data we finally reached a diagnosis of poorly resolving palsy of the posterior interosseous nerve. Based on this diagnosis, an operation was performed, exposing the radial nerve. It revealed soft tissue adhesions within the arcade of Frohse which were compressing the radial nerve. Detachment of the area of adhesion was performed and the compressing tissue removed. Extension of all fingers was confirmed by direct electrical stimulation of the radial nerve just after the operative
H. Kamo et al. / Clinical Neurology and Neurosurgery 110 (2008) 973–978
977
Fig. 6. A clinical case of the posterior interosseous nerve syndrome. (A) Photograph displaying the signs observed in case 2. Notice that the patient can still extend his wrist. (B) The responsible lesion was narrowed to the restricted portion of the posterior interosseous nerve near the arcade of Frohse after elimination of innervation to the muscle of ECR. (a) Posterior interosseous nerve (b) nerve to muscle of extensor carpi radialis brevis (ECR) (c) nerve to muscle of extensor digitorum communis (EDC) (d) nerve to muscle of extensor carpi ulnaris (ECU) (e) nerve to muscle of extensor pollicis longus (EPL).
procedure. Remarkable recovery was observed several months after the operation. 3.2.3. Case 3, a sensory PNS lesion A 70-year-old man developed a sensory disturbance from the right shoulder to the right thumb. Numbness, sometimes with dull pain, continued throughout the day. It worsened when lying on the right side and when he pulled on his dog’s leash while walking. He was a keen table tennis player and continued to practice his swing every day. He attended a nearby orthopedic clinic, and had been undergoing head traction for several months following a diagnosis of cervical spondylosis. However, his symptoms worsened, and he consequently visited this hospital. Neurological examination revealed an area of sensory disturbance in the region of the right arm shown in green in Fig. 7A. No abnormalities were found in his tendon reflexes, muscle strength, higher neurological function or his autonomic nervous system. There was neither ataxia nor rigidity. No abnormal findings were detected by sphygmography. Chest X-ray examination revealed deformity of the right upper thorax. Although cervical spondylosis of C4, C5 and C6 was observed on X-ray images, no stenosis of the intervertebral canal was evident. Cervical MRI showed disc bulging from C3/4 to C7/T1, but no canal stenosis. There were no abnormalities on cerebral MRI or on blood examination including Hb A1C, thyroid hormone and amyloid A protein. Computer display of the disturbed sensory region is shown as the narrowed red line indicating the overlapping pathways in the brachial plexus that could be responsible (Fig. 7B). EMG tests showed normal conduction in the median, ulnar and radial nerves. However, decreased F-wave frequency was found in the right radial nerve. Needle EMG of the right deltoid and triceps muscles showed abnormal recruitment and/or unit morphology. These tests were normal when applied to the biceps brachii, rhomboid, serratus anterior and infraspinatus muscles. From the results above, EMG tests indicated that the posterior cord was the most plausible site of the brachial plexus lesion. Rest cure was prescribed after the diagnosis and the patient’s condition resolved completely after several months.
4. Discussion Any approach to neurological diagnosis based on observed neurological deficits needs to have a logical foundation. The underlying goal is to identify a single anatomical site where a lesion could produce the observed deficits while sparing other systems. These simple concepts should apply even in refractory disorders such as multiple sclerosis where multiple lesions may occur [1]. Since this preliminary investigation was undertaken only to assess the usefulness of a computer system based on anatomical features of a typical human body, we did not take many possible refinements into consideration. These could include individual differences in the dermatome [14,16–19] and myotome [14,18,20,21]. Nevertheless, the illustrations provided here show that information stored on computer, when combined with systematic methods of recall, make it possible to pinpoint lesions that otherwise might have been missed or overlooked. Using this system we confirmed that a computer program can support and enhance routine clinical diagnosis through its large memory capacity, accuracy, and fast recall. Despite remarkable progress in the technology used in diagnostic imaging in the past few decades, for example X-ray, ultrasound, CT, MRI, and PET, the user needs to have experience and knowledge in order to use these tools effectively. The level of knowledge of the user decides the quantity and quality of the information obtained. Precise knowledge of neurological anatomy is particularly important in neurological anatomical diagnosis, but because of the complexity of the anatomy accurate memorization is difficult. Applying the same logic used in everyday neurological diagnosis – localizing a lesion based on symptoms and knowledge of neurological pathways – the system has proved its ability when applied to real-world clinical data. Based on these results we conclude that this system can both depict abnormalities in neurological pathways based on patients’ symptoms and complement more usual diagnostic imaging using MRI and CT. The system we have presented here is a novel imaging technology using a new perspective based on advances in computer
978
H. Kamo et al. / Clinical Neurology and Neurosurgery 110 (2008) 973–978
Fig. 7. A clinical case of brachial plexopathy. (A) The region of sensory disturbance in case 3 (green area). (B) Responsible pathways corresponding to each subdivided area overlapped primarily at a part of brachial plexus (a). (b) Lateral cord of brachial plexus (c) anterior division of brachial plexus.
technology. Later developments might include 3D visualization and direct superimposition on patient MRIs. Advances in areas such as neurological anatomy will produce vast amounts of information. By integrating this information and advances in imaging technology, this system will have the potential to evolve into a system able to help the user gain a higher level of knowledge in the field of neurology. Future research will probably produce computer software programs with more precise and abundant anatomical information, and development of these systems will probably result in more refined results. Despite the requirement of the computer program that results always need to be controlled by a clinician, this system might lead to an incorrect diagnosis owing to erroneous input of information. Correct input of information requires exact identification of clinical deficits affecting muscles and nerves. As this system simply provides computer assistance in finding or predicting anatomical lesions within the CNS and PNS, final diagnosis should take all variables including patient characteristics and inter-individual differences into account. We believe that despite the limitations we have mentioned, this system is particularly useful for cases in which many logical inferences are required in the diagnostic process, or for cases with spatially complicated integration, for instance, cases in which plexus or brain stem disorders are encountered, as it may facilitate diagnosis of such cases. This support system can even be used by experienced neurologists to check, extend or complete their knowledge. Since human memory has its limitations, to us it seems inevitable that in the future computers will be used by neurologists, neurosurgeons, radiologists and many others, perhaps as illustrated here. References [1] Montgomery EB, Wall M, Henderson VW. Principles of neurologic diagnosis. 1st ed. Boston: Little, Brown and Company; 1986. p. 4–7.
[2] Haymaker W. Bing’s local diagnosis in neurological diseases. 15th ed. Saint Louis: The C.V. Mosby Company; 1969. p. 217–22. [3] Kimura J. Electrodiagnosis in diseases of nerve and muscle: principles and practice. 3rd ed. New York: Oxford University Press; 2001. p. 3–26, 628– 49. [4] Fisher WS. Computer aided intelligence: application of an Expert System to brachial plexus injuries. Neurosurgery 1990;27:837–43. [5] First MB, Weimer BJ, Mc Linden S, et al. Computer-assisted localization of peripheral nervous system lesions. Comput Biomed Res 1982;15:525–43. [6] Hudgins WR. Computer-aided diagnosis of lumbar disc herniation. Spine 1983;8:604–15. [7] Mathew B, Norris D, Hendry D, et al. Artificial intelligence in the diagnosis of low-back pain and sciatica. Spine 1988;13:168–72. [8] Meyer AU, Weissman WK. Computer analysis of the clinical neurological examination. Comput Biol Med 1973;3:111–7. [9] Reggia JA. A production rule system for neurological localization. In: Proceedings, second annual symposium on computer applications in medical care. Washington, DC: IEEE Computer Soc.; 1978. p. 254–60. [10] Catanzarite VA, Greenburg AG, Bremermann HJ. Computer consultation in neurology: subjective and objective evaluations of the “Neurologist” system. Comput Biol Med 1982;12:343–55. [11] Banks G, Weimer B. Symbolic coordinate anatomy for neurology (SCAN). J Med Syst 1984;8:157–62. [12] Spitzer K, Thie A, Caplan LR, et al. The TOPSCOUT expert system for stroke localization. Stroke 1989;20:1195–201. [13] DeArmond SJ, Fusco MM, Dewy MM. Structure of the human brain, a photographic atlas. 3rd ed. New York: Oxford University Press; 1989. p. 63–143. [14] Haymaker W, Woodhall B. Peripheral nerve injuries. Principles of diagnosis. 2nd ed. Philadelphia: Saunders; 1953. p. 17–47, 222–72. [15] Hislop HJ, Montgomery J. Daniels and Worthingham’s muscle testing techniques of manual examination. 7th ed. Philadelphia: Saunders; 2002. p. 14–252. [16] Foerster O. Dermatomes in man. Brain 1933;56:1–39. [17] Keegan JJ, Garrett FD. The segmental distribution of the cutaneous nerves in the limbs of man. Anat Record 1948;102:409–37. [18] O’ Brien MD. Aids to the examination of the peripheral nervous system. 4th ed. Edinburgh: W.B. Saunders; 2000. p. 56–62. [19] Ropper AH, Brown RH. Adams and Victor’s principle of neurology. 8th ed. New York: McGraw-Hill; 2005. p. 129–43. [20] Liguori R, Krarup C, Trojaborg W. Determination of the segmental sensory and motor innervation of the lumbosacral spinal nerves: an electrophysiological study. Brain 1992;115:915–34. [21] Levin KH, Maggiano HJ, Wilbourn AJ. Cervical radiculopathies: comparison of surgical and EMG localization of single-root lesions. Neurology 1996;46:1022–5.
Contents lists available at ScienceDirect
Clinical Neurology and Neurosurgery journal homepage: www.elsevier.com/locate/clineuro
A computer support system for neurological anatomical diagnosis Hisaki Kamo a,∗ , Yosihisa Kiriyama b,1 , Akihiro Mizoe b,1 , Eiko Murase c,2 , Seiichiro Okajima d,3 , Ichiro Akiguchi e,4 , Yasusuke Hirasawa f,5 , Patrick L. McGeer g,6 a
Department of Neurology, Uzumasa Hospital, 30 Katabiranotujimachi, Uzumasa, Ukyo-Ku, Kyoto City 616-8151, Japan System Laboratory Murata Co., Ltd., 18-8 Kinoshita-cho, Otsu City, Shiga Prefecture 520-0812, Japan c Department of Neurology, Utano National Hospital Organization, 8 Ondoyama-cho, Narutaki, Ukyo-ku, Kyoto City 616-8255, Japan d Department of Orthopedics, Shakaihoken Kobe Central Hospital, 1-1, 2 choume, Souzan-chou, Kita-ku, Koube City, Hyougo Prefecture, Japan e Department of Neurology, Koseikai Takeda Hospital, 841-5, Higashishionokouji-cho, Higashiiru, Shionokoji-dori Nishinotouin, Shimokyou-ku, Kyoto City 600-8558, Japan f Graduate School, Meiji University of Integrative Medicine, Hiyoshi-cho, Funai-gun, Kyoto City 629-0392, Japan g Kinsmen Laboratory of Neurological Research, University of British Columbia, 2255 Wesbrook Mall, Vancouver, BC, Canada V6T 1Z3 b
a r t i c l e
i n f o
Article history: Received 25 January 2008 Received in revised form 10 April 2008 Accepted 27 May 2008 Keywords: Neurological diagnosis Neuroanatomy Anatomical diagnosis Computer Central nervous system Peripheral nervous system Automatic depicting system
a b s t r a c t Objectives: Accurate neurological diagnoses are often difficult to make due to the complexity of the neuroanatomy involved. This study was performed to evaluate the usefulness of a computer system with easily retrievable anatomical information as a support for arriving at more accurate anatomic diagnoses. Patients and methods: Anatomical information from an initial physical examination was programmed into a computer with stored neuroanatomical charts of the brain, spinal cord and peripheral nerves. The information generated a graphic display of possible lesions with suggestions for further examination. These suggestions were then followed and further data entered. This data entry generated a new graphic display with reduced lesion possibilities. Iterations were then followed to narrow the possibilities for diagnosis further, until a final anatomical diagnosis was reached. This method was applied to three hypothetical examples and a number of clinical cases. Here we report three clinical cases in which this method was particularly useful in making a diagnosis. Results: Using computer iterations, the system was able to pinpoint one or more presumptive causative lesions in the CNS or PNS based on known neuronal pathways or nuclei. Conclusion: The results indicate that suitably used, computer memory, by virtue of its large capacity, accuracy and fast recall, can supplement human memory in reaching accurate anatomical diagnoses of neurological lesions. © 2008 Elsevier B.V. All rights reserved.
1. Introduction Accurate diagnoses of neurological diseases can be difficult to make due to the complexity of human neuroanatomy. Furthermore, the problem can be compounded when multiple causative neuronal pathways and nuclei are involved. While magnetic resonance imaging (MRI) and computed tomography (CT) images are frequently relied upon for accurate diagnostic
∗ Corresponding author. Tel.: +81 75 871 0505. E-mail address: [email protected] (H. Kamo). 1 Tel.: +81 77 521 7020. 2 Tel.: +81 75 461 5121. 3 Tel.: +81 78 594 2211. 4 Tel.: +81 75 361 1351. 5 Tel.: +81 771 72 1181. 6 Tel.: +1 604 822 7377. 0303-8467/$ – see front matter © 2008 Elsevier B.V. All rights reserved. doi:10.1016/j.clineuro.2008.05.022
purposes, they may reveal multiple abnormalities making pinpointing of the actual causative lesion difficult. Pattern recognition of neurological signs and symptoms is another approach to accurate diagnosis. Although this approach might seem straightforward, it actually requires a very accurate recall of many specific anatomical details related to numerous disease entities. There may also be significant difficulties when dealing with unusual diseases or unusual manifestations of common diseases [1]. We have studied the possibility of using computer assistance to map the neuronal pathways involved in a patient’s neurological symptoms as a means of accurately localizing the lesions and thus contributing to accurate diagnoses [1–3]. A computer software program was generated in which neuroanatomical charts of the brain, spinal cord and peripheral nerves were stored. Specific charts could then be displayed following input prompting from the clinician according to the physical examination of the patient.
974
H. Kamo et al. / Clinical Neurology and Neurosurgery 110 (2008) 973–978
The principal purpose of this study was to evaluate the ability of this computer program to assist the clinician to pinpoint lesions by displaying information consistent with the patient’s neurological symptoms. Several trials have been performed mostly on pattern recognition diagnoses in restricted areas of the PNS [4–7] and the CNS [8–12] such as the brachial plexus or the brain stem. These were mainly done in the 1970s and 1980s. However, a comprehensive system has not yet been developed for common usage. Advances in computer technology have now made it practical to develop such a system and a prototype version is presented here.
2. Methods A computer program was developed where neuroanatomical charts of the brain, spinal cord [13] and peripheral nerves [14,15] could be stored. Microsoft Visual C++ .NET 2003 and Access 2000 were used to produce all drawings. A depicting system was then programmed to display appropriate neuronal pathways following user input of the patient’s signs and symptoms. For example, in the peripheral motor system, the innervation of each muscle was stored from the nerve root to its distal end. The system was then programmed to display possible lesions corresponding to weaknesses or abnormal neurological findings from EMG test results of each muscle. For the sensory system, all areas of the body surface were subdivided into minimal cutaneous field units corresponding to the sensory spinal roots (dermatome) and peripheral nerves. The software was programmed to depict the responsible pathways, using colored shading from light blue to red to indicate the degree of overlap in cases with more than one symptom. In addition to the depicting systems described above, the software was written to eliminate intact parts of the motor system thus enabling the narrowing down of causative lesions [3,5]. For the central nervous system, the software was programmed to trace the distance along causative neuronal pathways or nuclei on 25 axial sections of the brain and spinal cord, and to indicate the perimeter area excluding all unrelated symptoms [1].
Fig. 1. Computer display following operator input of the symptoms of the Millard–Gubler syndrome. It displays in brown the minimal polygonal area in the lower pons that includes two corticospinal pathways, the facial nerve and the abducens nerve. (a) Facial nerve (b) abducens nerve (c,d) corticospinal tract to upper (c) and lower extremity (d).
3.1.2. Example 2, weakness involving the peripheral motor system In the peripheral motor system, neuronal pathways corresponding to any muscle weakness or EMG abnormality can be displayed. Overlapping pathways are demonstrated by differences in color corresponding to the extent of overlap. Neuronal pathways to normal muscles are shown as broken lines. For example, in a hypothetical case with weakness at the thumb interphalangeal (IP) joint and distal interphalangeal (DIP) joint of forefinger, but without involvement of the metacarpophalangeal (MCP) joint of the thumb (the so-called pinch sign in the anterior interosseous nerve syndrome), the system displays, as the causative lesion, that part of the arm between the start of interosseous nerve and the point of branching to the long flexor muscle of thumb. The unaffected nerve is shown by the broken line (Fig. 3).
3. Results 3.1. Hypothetical examples 3.1.1. Example 1, motor paralysis of CNS origin In a hypothetical examination, clinical symptoms of upper and lower right sided extremity paralysis were observed. The data were entered, and the system responded immediately by displaying the corresponding neuronal pathways on 25 sectional diagrams of the brain and spinal cord. Iteration through additional input was then carried out to allow the potential causative lesion to be deduced. For instance, if the case also had left subnuclear facial paralysis and left abducens palsy in addition to the extremity paralysis, these could be entered into the system described above. All the responsible neuronal pathways were then drawn on a single section of the pons. A polygonal area covering the lower and ventral pontine region corresponding to the Millard–Gubler syndrome was then displayed (Fig. 1). In other circumstances, several levels might be involved. Fig. 2, for example, illustrates the results of such a case. Multiple areas for a causative lesion which produces right motor paralysis plus impairment of thermal-pain and deep sensation are displayed.
3.1.3. Example 3, sensory disturbance in the peripheral sensory system For the peripheral sensory system, the computer displays the whole body showing the appropriate neuronal pathways, with graded coloring indicating the degree of overlap of innervation. In a hypothetical case where disturbed sensation from the thumb to the lateral forearm is presented, which is typical of a C6 radiculoneuropathy, three peripheral nerves might be involved. These are the radial nerve, the median nerve and the lateral antebrachial cutaneous nerve. These are displayed in Fig. 4A. Responsible pathways from the distal end of a nerve to its root are drawn by the subdivided method described above, and the root segment (arrow in Fig. 4B) is identified as the site of the causative lesion, following triple iteration. 3.2. Clinical case examples 3.2.1. Case 1, a CNS lesion A 51-year-old, right-handed man experienced sudden dysphagia, dysarthria and vomiting at a company meeting. He complained of numbness of the right hand, clumsiness of the left hand, and mild unsteadiness when walking. He went to see his family doc-
H. Kamo et al. / Clinical Neurology and Neurosurgery 110 (2008) 973–978
975
Fig. 3. An inferred example of the motor system of peripheral nerves with the anterior interosseous nerve syndrome. The intact median nerve is shown as a broken line and can be eliminated. (a) Median nerve. The affected area of the nerve is displayed as a solid line (b) anterior interosseous nerve (c) nerve to muscle of flexor pollicis longus (d) nerve to muscle of flexor digitorum profundus. Fig. 2. A case with multiple inferred areas. See Section 3 for details of symptoms. (A) Middle pons (B) lower pons (C) upper medulla oblongata. (a) Corticospinal tract (b) lateral spinothalamic tract (c) dorsal column – medial lemniscus pathway.
tor, accompanied by his colleagues, but no abnormal findings were detected by CT that day or by MRI several days later. Since no lesion was identified, a psychosomatic disorder or masked depression was suspected. Accordingly, antidepressant and tranquilizer drugs were prescribed. For several weeks there was only slight improvement in the dysphagia, dysarthria and gait disturbance. Facial and right leg numbness began to develop. As a result, and due to his anxiety, he visited our hospital. Neurological examination revealed the following positive findings: sensory disturbance of pain and temperature in the right upper and lower extremities and left face. There was no impairment of touch or deep sensation, but there was loss of the left gag reflex, mild pharyngeal dysarthria and dysdiadochokinesis of the left hand plus a slight gait ataxia. There was no weakness or rigidity of the upper and lower extremities.
Following entry of these findings into the computer, the causative lesion was demonstrated to be in the left lateral medulla oblongata corresponding to the Wallenberg syndrome (Fig. 5A). Routine MRI tests had been unable to detect the lesion. However, thin slice examination revealed a T1-low, T2-high area in the lateral medulla oblongata confirming the lesion identified by the computer software (Fig. 5B). 3.2.2. Case 2, a motor PNS lesion A 36-year-old man developed right elbow pain and was unable to extend all the fingers of his right hand but was still able to extend his wrist (Fig. 6A). He visited a nearby hospital to consult with an orthopedic surgeon. He was diagnosed as having radial nerve palsy, but due to an inability to localize the lesion precisely, he was treated conservatively for about 4 months. Since he experienced little improvement in his loss of motor function, he visited our hospital. In order to determine the level of the radial nerve lesion, the programmed software was used to anatomically simulate the muscle disturbance
976
H. Kamo et al. / Clinical Neurology and Neurosurgery 110 (2008) 973–978
Fig. 4. Computer display of a sensory nerve disturbance. (A) Area of sensory disturbance (red area) in C6 radiculopathy is displayed on the subdivided body surface with spinal dermatome (red line) and peripheral nerve innervation (blue line). Supply from (a) C6 and lateral antebrachial cutaneous nerve (b) C6 and radial nerve (c) C6 and median nerve (d) C7 and median nerve. (B) Corresponding neuronal pathways of the three divisions overlap at the portion close to the C6 root.
Fig. 5. A clinical case of the Wallenberg syndrome. (A) Computer display of the neuronal pathways corresponding to the symptoms of case 1. (a, b) Lateral spinothalamic tract from upper (a) and lower (b) extremities. (c) Spinal trigeminal tract (d) motor nerve from the ambiguous nucleus. (B) Thin slice MRI image of the medulla oblongata in case 1. A high intensity area was observed in the dorsolateral region (arrow) of the T2 image.
anticipated for lesions at various levels, prior to examining the patient. The muscles chosen were as follows: deltoid, triceps brachii, supinator, extensor carpi radialis longus and brevis (ECR), extensor carpi ulnaris (ECU), extensor pollicis longus (EPL), and the extensor digitorum communis (EDC). Manual muscle testing (MMT) of all these candidate muscles revealed that weakness was confined to the last four muscles. Especially at the wrist, the two carpal extension muscles were found to be particularly important according to the programmed software. MMT was therefore done in the radial direction and the ulnar direction respectively. The result was 5/5 in the radial direction but only 1/5 in the ulnar direction. The results were entered into the software program and the causative lesion was diagnosed as occurring in the upper part of the posterior interosseous nerve (Fig. 6B). EMG testing then demonstrated a denervation potential using fibrillation voltage or positive sharp wave in the resting state for
the muscles of ECU, EPL and EDC, but not for brachioradialis, ECR, deltoid, or triceps. Only minimal discharge could be found in the muscles of ECU, and no reinnervation potential was found in EPL and EDC following voluntary contraction. In conclusion, little tendency for recovery was shown in the electromyographic tests. SEP tests demonstrated normal findings and cervical MRI showed only mild disc bulging at C6/7 without canal or intervertebral canal stenosis. The result of sphygmography was normal. Using these data we finally reached a diagnosis of poorly resolving palsy of the posterior interosseous nerve. Based on this diagnosis, an operation was performed, exposing the radial nerve. It revealed soft tissue adhesions within the arcade of Frohse which were compressing the radial nerve. Detachment of the area of adhesion was performed and the compressing tissue removed. Extension of all fingers was confirmed by direct electrical stimulation of the radial nerve just after the operative
H. Kamo et al. / Clinical Neurology and Neurosurgery 110 (2008) 973–978
977
Fig. 6. A clinical case of the posterior interosseous nerve syndrome. (A) Photograph displaying the signs observed in case 2. Notice that the patient can still extend his wrist. (B) The responsible lesion was narrowed to the restricted portion of the posterior interosseous nerve near the arcade of Frohse after elimination of innervation to the muscle of ECR. (a) Posterior interosseous nerve (b) nerve to muscle of extensor carpi radialis brevis (ECR) (c) nerve to muscle of extensor digitorum communis (EDC) (d) nerve to muscle of extensor carpi ulnaris (ECU) (e) nerve to muscle of extensor pollicis longus (EPL).
procedure. Remarkable recovery was observed several months after the operation. 3.2.3. Case 3, a sensory PNS lesion A 70-year-old man developed a sensory disturbance from the right shoulder to the right thumb. Numbness, sometimes with dull pain, continued throughout the day. It worsened when lying on the right side and when he pulled on his dog’s leash while walking. He was a keen table tennis player and continued to practice his swing every day. He attended a nearby orthopedic clinic, and had been undergoing head traction for several months following a diagnosis of cervical spondylosis. However, his symptoms worsened, and he consequently visited this hospital. Neurological examination revealed an area of sensory disturbance in the region of the right arm shown in green in Fig. 7A. No abnormalities were found in his tendon reflexes, muscle strength, higher neurological function or his autonomic nervous system. There was neither ataxia nor rigidity. No abnormal findings were detected by sphygmography. Chest X-ray examination revealed deformity of the right upper thorax. Although cervical spondylosis of C4, C5 and C6 was observed on X-ray images, no stenosis of the intervertebral canal was evident. Cervical MRI showed disc bulging from C3/4 to C7/T1, but no canal stenosis. There were no abnormalities on cerebral MRI or on blood examination including Hb A1C, thyroid hormone and amyloid A protein. Computer display of the disturbed sensory region is shown as the narrowed red line indicating the overlapping pathways in the brachial plexus that could be responsible (Fig. 7B). EMG tests showed normal conduction in the median, ulnar and radial nerves. However, decreased F-wave frequency was found in the right radial nerve. Needle EMG of the right deltoid and triceps muscles showed abnormal recruitment and/or unit morphology. These tests were normal when applied to the biceps brachii, rhomboid, serratus anterior and infraspinatus muscles. From the results above, EMG tests indicated that the posterior cord was the most plausible site of the brachial plexus lesion. Rest cure was prescribed after the diagnosis and the patient’s condition resolved completely after several months.
4. Discussion Any approach to neurological diagnosis based on observed neurological deficits needs to have a logical foundation. The underlying goal is to identify a single anatomical site where a lesion could produce the observed deficits while sparing other systems. These simple concepts should apply even in refractory disorders such as multiple sclerosis where multiple lesions may occur [1]. Since this preliminary investigation was undertaken only to assess the usefulness of a computer system based on anatomical features of a typical human body, we did not take many possible refinements into consideration. These could include individual differences in the dermatome [14,16–19] and myotome [14,18,20,21]. Nevertheless, the illustrations provided here show that information stored on computer, when combined with systematic methods of recall, make it possible to pinpoint lesions that otherwise might have been missed or overlooked. Using this system we confirmed that a computer program can support and enhance routine clinical diagnosis through its large memory capacity, accuracy, and fast recall. Despite remarkable progress in the technology used in diagnostic imaging in the past few decades, for example X-ray, ultrasound, CT, MRI, and PET, the user needs to have experience and knowledge in order to use these tools effectively. The level of knowledge of the user decides the quantity and quality of the information obtained. Precise knowledge of neurological anatomy is particularly important in neurological anatomical diagnosis, but because of the complexity of the anatomy accurate memorization is difficult. Applying the same logic used in everyday neurological diagnosis – localizing a lesion based on symptoms and knowledge of neurological pathways – the system has proved its ability when applied to real-world clinical data. Based on these results we conclude that this system can both depict abnormalities in neurological pathways based on patients’ symptoms and complement more usual diagnostic imaging using MRI and CT. The system we have presented here is a novel imaging technology using a new perspective based on advances in computer
978
H. Kamo et al. / Clinical Neurology and Neurosurgery 110 (2008) 973–978
Fig. 7. A clinical case of brachial plexopathy. (A) The region of sensory disturbance in case 3 (green area). (B) Responsible pathways corresponding to each subdivided area overlapped primarily at a part of brachial plexus (a). (b) Lateral cord of brachial plexus (c) anterior division of brachial plexus.
technology. Later developments might include 3D visualization and direct superimposition on patient MRIs. Advances in areas such as neurological anatomy will produce vast amounts of information. By integrating this information and advances in imaging technology, this system will have the potential to evolve into a system able to help the user gain a higher level of knowledge in the field of neurology. Future research will probably produce computer software programs with more precise and abundant anatomical information, and development of these systems will probably result in more refined results. Despite the requirement of the computer program that results always need to be controlled by a clinician, this system might lead to an incorrect diagnosis owing to erroneous input of information. Correct input of information requires exact identification of clinical deficits affecting muscles and nerves. As this system simply provides computer assistance in finding or predicting anatomical lesions within the CNS and PNS, final diagnosis should take all variables including patient characteristics and inter-individual differences into account. We believe that despite the limitations we have mentioned, this system is particularly useful for cases in which many logical inferences are required in the diagnostic process, or for cases with spatially complicated integration, for instance, cases in which plexus or brain stem disorders are encountered, as it may facilitate diagnosis of such cases. This support system can even be used by experienced neurologists to check, extend or complete their knowledge. Since human memory has its limitations, to us it seems inevitable that in the future computers will be used by neurologists, neurosurgeons, radiologists and many others, perhaps as illustrated here. References [1] Montgomery EB, Wall M, Henderson VW. Principles of neurologic diagnosis. 1st ed. Boston: Little, Brown and Company; 1986. p. 4–7.
[2] Haymaker W. Bing’s local diagnosis in neurological diseases. 15th ed. Saint Louis: The C.V. Mosby Company; 1969. p. 217–22. [3] Kimura J. Electrodiagnosis in diseases of nerve and muscle: principles and practice. 3rd ed. New York: Oxford University Press; 2001. p. 3–26, 628– 49. [4] Fisher WS. Computer aided intelligence: application of an Expert System to brachial plexus injuries. Neurosurgery 1990;27:837–43. [5] First MB, Weimer BJ, Mc Linden S, et al. Computer-assisted localization of peripheral nervous system lesions. Comput Biomed Res 1982;15:525–43. [6] Hudgins WR. Computer-aided diagnosis of lumbar disc herniation. Spine 1983;8:604–15. [7] Mathew B, Norris D, Hendry D, et al. Artificial intelligence in the diagnosis of low-back pain and sciatica. Spine 1988;13:168–72. [8] Meyer AU, Weissman WK. Computer analysis of the clinical neurological examination. Comput Biol Med 1973;3:111–7. [9] Reggia JA. A production rule system for neurological localization. In: Proceedings, second annual symposium on computer applications in medical care. Washington, DC: IEEE Computer Soc.; 1978. p. 254–60. [10] Catanzarite VA, Greenburg AG, Bremermann HJ. Computer consultation in neurology: subjective and objective evaluations of the “Neurologist” system. Comput Biol Med 1982;12:343–55. [11] Banks G, Weimer B. Symbolic coordinate anatomy for neurology (SCAN). J Med Syst 1984;8:157–62. [12] Spitzer K, Thie A, Caplan LR, et al. The TOPSCOUT expert system for stroke localization. Stroke 1989;20:1195–201. [13] DeArmond SJ, Fusco MM, Dewy MM. Structure of the human brain, a photographic atlas. 3rd ed. New York: Oxford University Press; 1989. p. 63–143. [14] Haymaker W, Woodhall B. Peripheral nerve injuries. Principles of diagnosis. 2nd ed. Philadelphia: Saunders; 1953. p. 17–47, 222–72. [15] Hislop HJ, Montgomery J. Daniels and Worthingham’s muscle testing techniques of manual examination. 7th ed. Philadelphia: Saunders; 2002. p. 14–252. [16] Foerster O. Dermatomes in man. Brain 1933;56:1–39. [17] Keegan JJ, Garrett FD. The segmental distribution of the cutaneous nerves in the limbs of man. Anat Record 1948;102:409–37. [18] O’ Brien MD. Aids to the examination of the peripheral nervous system. 4th ed. Edinburgh: W.B. Saunders; 2000. p. 56–62. [19] Ropper AH, Brown RH. Adams and Victor’s principle of neurology. 8th ed. New York: McGraw-Hill; 2005. p. 129–43. [20] Liguori R, Krarup C, Trojaborg W. Determination of the segmental sensory and motor innervation of the lumbosacral spinal nerves: an electrophysiological study. Brain 1992;115:915–34. [21] Levin KH, Maggiano HJ, Wilbourn AJ. Cervical radiculopathies: comparison of surgical and EMG localization of single-root lesions. Neurology 1996;46:1022–5.