(1)
Department of Neurology, Wake Forest University School of Medicine, Winston-Salem, NC, USA
Abstract
Medical diagnosis and treatment decision making traditionally hinges on clinical judgment, available evidence from the literature and “test results”. This information is highly imperfect and can be erroneous or frankly misleading. This chapter borrows from digital fly-by wire paradigms for processing information from unreliable sources for guiding diagnosis and treatment. A brief overview of the principles and architectures used in Airbus and Boeing Fly-By Wire systems is presented. The relevant architectures are used to inspire similar methods of thinking for medical diagnosis and treatment. This chapter extends the theory of fault tree analysis and probabilistic methods developed in prior chapters for guiding decision making in the face of uncertainty and misleading clinical information. Costs and treatment risks are incorporated into uncertainty associated with poor quality information. Medical examples are presented.
Keywords
Byzantine generals problemBoeing fly-by wireAirbus fly-by wireMisleading informationDistal myopathy with rimmed vacuolesSubacute combined degeneration of the cordVitamin B12 deficiencyLambert Eaton Myasthenic Syndrome (LEMS)Chronic inflammatory demyelinating polyneuropathyAcute inflammatory demyelinating polyneuropathyByzantine Faults
A famous problem in computer science is the Byzantine Generals problem [1]. In this hypothetical situation, two or more byzantine generals need to attack an enemy fort. Each does not have the strength to attack alone and only a coordinated attack will succeed. They must coordinate and communicate their attack plans through unreliable, potentially treacherous messengers who can fail to deliver the message, can distort the message to mislead the other army or be intercepted by the enemy. Similarly in the dependability literature, byzantine faults refers to unpredictable system failures where failures manifest not merely as lack of output but as unpredictable and misleading system output. As shown in Fig. 1.6 of Chap. 1, such faults can cascade across systems with potentially catastrophic consequences.
Fault tolerant systems must deal with Byzantine faults [1]. To a great extent, medical decision making and treatment must deal with byzantine faults. Patient-reported symptoms, physical examination findings, laboratory findings, radiological findings, and neurophysiology (EEG, EMG) findings are prone to errors in understanding and interpretation which can mislead decision making with deleterious consequences. While there is no solution to the Byzantine generals problem, approximate solutions are used in day-to-day applications to deliver dependable service [1].
In traditional passenger airplanes, mechanical linkages in the form of hydraulics are used to transmit inputs from the pilot to movable sections of wings, horizontal stabilizers (elevators), rudder, called control surfaces to fly the airplane. This is shown in Fig. 6.1. The first analogue electrical flight control system, Fly-By-Wire (FBW), for a civil aircraft was designed by Aerospatiale for the Concorde [2]. In fly-by-wire systems, mechanical linkages are replaced by electrical wires which move the control surfaces using actuators (an actuator is a mechanism, usually electrohydraulic that physically moves the control surface) to control the airplane. The most exacting fault tolerant architectures which deal with byzantine faults are used in digital fly-by wire (DFBW) systems. This chapter briefly presents the chief features of DFBW used in Boeing and Airbus airplanes to describe two major architectures which are fault tolerant and capable of making safe decisions in the face of Byzantine faults [2–5].
Fig. 6.1
The Boeing 777 control surfaces. Slats and Flaps increase the curvature of the wing to increase lift during take-off and landing. The rudder moves the plane in the “yaw” axis (moves the nose of the plane right or left). The ailerons are used to move the airplane in the “roll axis”, where it rotates along its length to raise one wing and dip the other. The elevators and horizontal stabilizers move the plane in “pitch” axis where the nose goes up or down. © [1998] IEEE. Reprinted, with permission, from [4]
In DFBW airplanes, pilot inputs are interpreted by flight control computers to make the necessary movement of control surfaces to make desired flight path modifications [2]. In autopilot mode, the flight control computers take their orders from the autopilot flight director computers (AFDC). The heart of the fly-by wire system is the primary flight computers (PFCs). They receive inputs from the pilot flying the aircraft, information from sensors (such as ADIRU, accelerometers, and rate-gyro) and perform complex signal processing on these data [2–5]. The Air Data and Inertial Reference Unit (ADIRU) perform two critical functions: providing air data (airspeed, attitude, altitude) and inertial navigation information to the PFCs and pilots on the electronic flight display. The PFCs use these inputs to calculate control laws to compute the control surface position commands. These commands are then transmitted to the control surfaces using an electrical data bus. To meet extremely high functional integrity and functional availability requirements, multiple redundant hardware resources are required. FAR/CS 25. 1309 standards specify that any combination of failures of this system which can cause catastrophic consequences be “extremely improbable” with probability of occurrence less than 1 in 10 billion per flight hour of operation (see Chap. 1). The fault tolerance for trustworthy FBW system design should consider all known and unknown causes of problems, failures, and errors known as common mode failure and single point failure [4]. Based on these principles, the following sections look at the architectures used in Boeing and Airbus airplanes. Less technically inclined readers can skip sections “The Boeing 777/787 FBW Computers” and “Airbus Fly-By-Wire” and resume at section “Lessons from Digital Fly-By Wire”.
The Boeing 777/787 FBW Computers
The Boeing 777, which debuted in 1995 is the first commercial DFBW airplane manufactured by Boeing. This was followed by the 787 in 2011 with common principles governing the design of the flight control system. The key principles of the Boeing 777/787 flight control design philosophy are [5]:
(a)
The automation is an aid but does not replace the pilot.
(b)
The pilot has the highest authority.
(c)
The pilot and copilot must be aware of each other’s input.
(d)
Control functions will assist the pilot in avoiding or recovering from exceeding operational boundaries.
(e)
Control laws reduce pilot workload and improve ride quality.
(f)
Reliable alternate control mechanisms will be available to deal with failures.
The 777 uses a triple–triple modular redundancy design (TMR) for achieving fault tolerance [4]. TMR was introduced in Chap. 1 as a means of achieving fault tolerance.
The triple–triple modular system uses three identical channels termed left, center, and right PFC for redundancy. The outputs are transmitted to the control surfaces using three redundant ARINC (Aeronautical Radio, Inc.) 629 buses. There are three dissimilar computing lanes within each channel as shown in Fig. 6.2.
Fig. 6.2
The Boeing 777 Primary Flight Computers (PFC’s). The DFBW system consists of three identical channels (left, center, and right), each of which is composed of three dissimilar computing lanes powered by different microprocessors. Each PFC receives and transmits on one ARINC 629 bus but receives all three bus lanes. Adapted from [4, 5]
The PFCs are designed to comply with the following safety requirements:
(a)
Each PFC has three dissimilar computing lanes with three different processors using the ADA programming language. This is shown in Fig. 6.2. The three processors are the AMD 29050, Motorola 68040, and Intel 486 DX4 [4, 5]. As discussed in Chap. 1, failures can occur when certain input patterns combine with latent system faults and states to produce erroneous, unstable outputs. Using dissimilar processors manufactured by different companies in parallel mitigates this to a great extent since they are unlikely to fail together when presented with the same challenge, thereby mitigating against byzantine faults in the most complex component (microprocessor) of the system [4, 5].
Each PFC channel transmits on a preassigned data bus and receives on all the buses. This prevents one bad channel from disrupting all the communications. The three PFC channels are placed in different locations of the aircraft to prevent damage from structural causes from causing service failure of all systems.
In normal operation, the PFCs exchange information with each other for critical variable equalization to maintain convergence between PFC channel outputs. Each PFC lane operates in two modes—command role or monitor role. Only one lane in each channel is allowed to be in command mode [4, 5]. The PFC lanes in monitor role will perform a “selected output” monitoring of their command lane. If an error is detected in the command lane by the monitor lanes of a particular channel, it is declared bad and taken offline [4–6]. One of the spare lanes is upgraded to command assignment. The PFCs perform their calculations and exchange their proposed surface command outputs with each other. The proposed output from each channel is voted by a median value select algorithm in PFC hardware to produce the selected PFC command output. They declare the selected median value as the actual computed control value. The command lane will send the selected actual surface command to its ARINC 629 bus [4–6]. This prevents structural fatigue from force fights induced by asynchronous PFC channel operation [4, 5].
The median value select method provides fault blocking against PFC faults until completion of fault detection, identification, and reconfiguration via PFC cross lane monitoring [4, 5].
Therefore, nine distinct computing lanes organized in three PFC channels, using three different sets of microprocessors communicating via three different ARINC 629 data buses provide redundancy in calculating DFBW output. This prevents a byzantine fault in any one lane from producing erroneous output compromising stable flight [4, 5]. Since their debut in 1995, excellent in-service reliability, six times better than predicted has been achieved. No unsafe events have occurred in over 20 million flight hours with more than 1,000 airplanes in service [5].
Airbus Fly-By-Wire
The Airbus A320/A330/A340/A380 and soon to debut A350 XWB family of aircraft are all DFBW. The Airbus A320 which entered service in 1988 is the first of the current generation of DFBW airplanes [2]. The striking feature of the entire Airbus family of airplanes is that control laws electrically drive all the control surfaces; the pilot sets objectives and not directly a control surface position [2, 3]. In all Airbus planes, conventional control columns are replaced by sidesticks. Therefore control surface positions are the sum of pilot inputs and stabilization orders [3]. As stated previously, the flight control computers take their orders from the pilot or from the AFDC. The flight control computers consist of five to seven computers and the autopilot system of two. The Airbus flight control system supports four main functions [2, 3]:
(a)
Acquisition and monitoring of crew requests through sidesticks and associated sensors.
(b)
Acquisition and monitoring of aircraft response.
(c)
Piloting the aircraft via control laws so that the aircraft achieves the objectives set by the crew.
(d)
Control of actuators so that control surface position appropriately changes the aircraft position.
To meet extreme dependability and safety requirements, the flight control computers should not produce an erroneous signal. The basic element of the Airbus architecture is the command (COM) and monitoring (MON) failsafe computer [2, 3]. This is shown in Fig. 6.3. These computers are subject to draconian safety requirements and are functionally composed of a command channel and a monitoring channel. The system incorporates a high degree of redundancy to deliver fault tolerance.
The command channel ensures the function allocated to the computer (control of a mobile surface). The monitoring channel ensures the proper functioning of the command channel. These two computers have different functions and software. Both computers are simultaneously active [2, 3]. If the monitor computer senses an error and deselects the command order, it disables a solenoid valve on the corresponding actuator and it goes into stand-by mode. Two types of computers are used in the A320 FBW system: the ELAC (elevator and aileron computers) and the SEC (spoiler and elevator computers). Each computer includes a command channel and a monitoring channel. Therefore four different systems are used: the ELAC command channel, ELAC monitoring channel, SEC command, and SEC monitoring channel. This leads to four different types of software for redundancy [2, 3]. On the A320, two other computers are used for rudder control, the flight augmentation computers (FAC). The A320 has two ELAC computers, three SEC computers, and two FAC computers for redundancy.
On other DFBW Airbus planes (A330/A340 and A380), two different computers are used called PRIM (primary computers) and SEC (secondary computers). On these planes, rudder control is integrated into the PRIMs and SECs [2, 3].
Each channel (whether command or monitoring) includes one or more processors, associated memories, input and output circuitry, power supply and specific software [2, 3]. Each computer processes its inputs using different processors and software to avoid common mode faults. When the results of one of these two channels diverge sufficiently, the channel that detected the failure cuts the connection between the computer and the outside. The failure detection is achieved by comparing the difference between the command and monitoring orders with respect to a given threshold [2, 3]. This scheme allows the detection of consequences of a computer component failure and prevents the spread of the resultant error outside of the computer. Error detection is supplemented by monitoring the correct execution of the program and encoding of memories [2, 3].
Redundancy is managed at the system level. System functions are divided among all the computers so that each is constantly active on at least a subset of its functions [3]. For a given function, a computer is active, others are on stand-by (hot spares). As soon as the active computer interrupts its functioning, one of the computers in stand-by mode passes almost instantly into active mode with only a minimum perturbation of the control surfaces. These computers constantly transmit a signal of good health, when a failure is detected the signal is discontinued at the same time as the functional outputs to the actuator [3].
In the A340-600, four command and monitoring computers are used, with one being sufficient to power the aircraft [3]. In normal operation, one of the computers (PRIM1) provides control of the surface. The other computers control the other control surfaces. If PRIM1 or one of the actuators it controls fails, PRIM2 picks up the relay. Following the same type of failure, PRIM2 can pass on the task to SEC1 and then eventually to SEC2. In addition to redundancy, the system is capable of self-diagnosis. The DFBW system uses sensors distributed throughout the aircraft to sense failure and perform diagnosis.
Control laws need to be reconfigured in case of loss of some of the sensors, especially the ADIRU. For redundancy, each airplane has three ADIRUs. If all three ADIRUs are available as in the normal case, the pilot has full authority within a safe flight envelope since PFCs are able to make decisions with a high degree of accuracy and corroboration. If there are system failures and only 1 ADIRU is available, it is partly monitored by comparison to other independent sources of information. If all ADIRUs are lost, the protections provided by the PFCs are lost and control law finds itself in a limited degraded mode called direct mode which allows control similar to a conventional aircraft [2, 3]. The Airbus family of aircraft has accumulated a rich, extremely dependable experience in service [2, 3] with over 4,000 airplanes in service and over 80 million hours flown.
Lessons from Digital Fly-By Wire
Both Boeing and Airbus DFBW systems are extremely dependable and safe and have become the standard in civil aviation. The previous sections showed the greatly different systems used to achieve dependability. There are invaluable lessons for decision making to be learnt from the design principles of these systems which we will later extend to medical decision making.
1.
Information processing by the PFCs assumes Byzantine faults at all levels. The PFCs attempt to verify and corroborate information from multiple different sources (ADIRU’s, accelerometers etc.), so that no single or combination of byzantine faults can mislead the PFCs into making unsafe outputs. Therefore, under most circumstances, a faulty speed reading from a sensor cannot fool the PFC since it will be verified with data from a second and a third ADIRU prior to decisions being made.
2.
The Airbus and Boeing systems use enormous diverse redundancy using different computers driven by different microprocessors and software to make decisions. This mitigates against byzantine faults in hardware and software which prevents a single hidden flaw in the system from making unsafe outputs.
Therefore as hypothesized by Dr. Dionysius Lardner in 1834, reproduced in Chap. 1, Airbus and Boeing FBW systems embody the principle that the most certain and effectual check upon errors which arise in the process of computation, is to cause the same computations to be made by separate and independent computers; and this check is rendered still more decisive if they make their computations by different methods.
In medicine, frequently such an approach is not made. A single error in one medical test or a single misdiagnosis will act as the basis for decision making downstream causing error propagation and system failure as shown in Chap. 1. Applying DFBW principles learnt above can prevent or mitigate against such byzantine medical errors. Borrowing from DFBW, assume the following decision making principles:
(a)
All medical information whether they are patient-reported symptoms, physical examination findings, diverse test results, and radiology images should be considered unreliable and potentially misleading to varying degrees. The degree of unreliability varies from test to test: for example needle biopsies are highly fraught with sampling error, whereas large masses after they are resected are much less vulnerable to this susceptibility but still liable to errors in interpretation. This must be factored into medical decision making.
(b)
In complex medical cases, an important diagnosis should not be susceptible to single point failure—in other words, the assumptions and test results for a diagnosis should be clearly analyzed and not be vulnerable to a single point of failure. As seen in case Example 2, Chap. 3, a weakly positive AchR-binding antibody titer should not form the absolute basis for a diagnosis of myasthenia gravis and be used to direct $50,000 worth of unnecessary treatment.
(c)
Competing hypotheses must be entertained throughout the lifecycle of complex cases where a firm diagnosis is difficult to establish. In such cases, there should be constant striving to differentiate between the current working diagnosis and competing hypothesis. Fault Tree Analysis, Bayesian methods described in Chaps. 3 and 5 are some tools which can generate competing hypotheses and enable their continuous refinement and discrimination.
(d)
Borrowing from these DFBW systems, each healthcare professional has latent (hidden) vulnerabilities in knowledge and judgment just as each microprocessor and its related software does (byzantine faults). These faults become activated when faced with specific unforeseen clinical challenges (just as a microprocessor and its software produce errors when faced with specific input patterns which expose vulnerabilities). In a manner similar to these systems, this can be overcome only by using knowledge redundancy—seeking the independent, unbiased opinion of colleagues and professionals (with different vulnerabilities) in a concurrent manner facilitating knowledge and professional teamwork. It is assumed these concurrent opinions (analogous to the control-monitor role in Airbus planes or nine independent computation lanes in Boeing FBW) will reach conclusions using “different, independent” methods. A decision where there is reasonable consensus between two experts inspires greater confidence than a divergent one. If there is disagreement, the degree of disagreement and the degree to which it alters treatment decision making must be factored. For example, if expert 1 feels a patient has CIDP, expert 2 feels patient has vasculitic neuropathy (control-monitor role), while there is disagreement, there is commonality in both experts agreeing this is an inflammatory neuropathic condition and is likely to respond to anti-inflammatory measures like steroids and IVIG. Therefore, to a great degree there is concurrence and treatment maybe initiated. In a hypothetical case where expert 1 feels a patient has CIDP and expert 2 feels it is a variant of diabetic neuropathy, there is considerable divergence between their opinions and the matter maybe put to vote with a third concurrent opinion (similar to Boeing FBW) and the median value or majority vote selected prior to initiating treatment with expensive options like IVIG. The emphasis is on concurrent opinion where a matter is discussed and understood concurrently rather than delays waiting for second and third opinions. If a concurrent opinion is not possible, then the same professional must evaluate a different diagnosis as a virtual second opinion and discriminate between the two.
(e)
Similar to flight control systems, medical diagnosis and treatment could be managed based on the degree of uncertainty in their foundations. The draconian standards applicable to the PFCs are not applicable to the in-flight entertainment system and food warming system. The in-flight entertainment system can produce output to the relevant screen even if there is some degree of uncertainty in the choice of movie selection. A similar paradigm can be applied to guide treatment in such circumstances as shown in Fig. 6.4.
Fig. 6.4
A paradigm for treatment decision making in the face of uncertainty in diagnostic accuracy. The risks and costs associated with different treatments can be weighed in terms of confidence that is placed on the accuracy of the diagnosis
Diagnostic Accuracy can be divided into three groups:
1.
Low: These include diagnosis made on:
(a)
Weak test results, such as antibodies positive in low titer, mild elevation of muscle enzymes where unequivocal proof has not been well established.
(b)
Symptoms, especially self-reported experience of pain, numbness which is difficult to measure using objective means. Examples include fibromyalgia, forms of complex regional pain syndrome.
In such cases, it is reasonable to embark on low cost and risk to intermediate cost and risk treatments as shown in Fig. 6.4. High treatment risk or cost is very likely not justified.
2.
Intermediate: These include diagnosis made based on more objective physical examination findings (such as muscle strength, deep tendon reflexes) and test data. In such cases, there is some reproducible mild to moderate abnormality on physical examination findings or radiological data, however there is still room for uncertainty. Examples include abnormalities on MRI Brain raising concerns for multiple sclerosis. This situation is frequently encountered with neurophysiological measurements such as EEG, EMG, and visual evoked potentials. These information sources are fraught with poor diagnostic criteria, subjective errors in interpretation, and mild abnormalities form the basis of diagnosis leading to a nonnegligible probability of misclassification with huge ramifications for treatment costs and risks. This situation is also frequently encountered with biopsies due to sampling errors. Two real examples:
(a)
GE is a 75-year-old female who suffered a right frontal hemorrhage several years ago requiring evacuation. She was subsequently admitted for waxing and waning mental status. Routine video EEG recording showed frequent abnormalities with possible seizures originating from her right cerebral hemisphere. She was started on phenytoin (later switched to carbamazepine) and levetiracetam (Keppra). Subsequent prolonged video EEG recordings were read as concerning for partial status epilepticus and the dose and treatment changed to carbamazepine since phenytoin and Keppra were not helping. Despite this, there was no improvement in her mental status. Based on frequent seizures on EEG, a recommendation was made to initiate propofol or midazolam as part of general anesthesia. Fortunately, prior to initiating general anesthesia, applying the Byzantine generals model, a concurrent review of her EEG was requested. Review of her EEG by an experienced epileptologist reported none of the prior EEG recordings were seizures. Therefore, there was substantial disagreement between two experts which exceeded the threshold for unequivocal diagnosis of seizures. Based on lack of corroboration of seizures, propofol or midazolam were not initiated and their costs and risks (general anesthesia and intubation) avoided. The patient did develop severe transaminitis from carbamazepine which was managed conservatively.
(b)
ND was a 54-year-old female who noticed a lump on the side of her tongue. The lump was not painful and showed a mild increase in size over 6 months. She was evaluated by an oral surgeon who did a biopsy from the edge of the lesion. This was reported as a benign lesion. No further workup was done over the next 6 months. Approximately 1 year later, it had started showing ulcerative features concerning for malignancy. Surgical excision proved it was carcinoma of the tongue, by when it had metastasized to the neck lymph nodes. Despite wide excision and radiation, the disease proved fatal within 2 years.
When a diagnosis is made with intermediate confidence in its veracity, it is reasonable to embark on low and intermediate cost and risk treatment as shown in Fig. 6.4.
3.
High accuracy: This includes those rare happy situations where multiple sources corroborate a diagnosis and consequent treatment. Examples include conclusive biopsies, pathognomonic CT/MRI scans where the imaging features are unambiguous. Neurological examples include biopsy proven nerve vasculitis. In some conditions like CIDP which are difficult to diagnose, diagnostic accuracy is increased by multiple features which are corroborative. For example, mild-to-moderate nerve abnormalities alone may have only intermediate diagnostic accuracy. However, when combined with the typical clinical picture of areflexia, symmetric proximal and distal weakness, high spinal fluid protein, the combination increases the diagnostic accuracy even though no one feature is conclusive in itself. This is analogous to the condition described in Chap. 1 where multiple lower development assurance level (DALs) can be used in lieu of a single higher DAL.
In such circumstances, it is reasonable to embark on treatments involving high costs and risks if needed since the treatment is based on high confidence in the diagnosis. The high confidence stems from the diagnosis being supported by more than one feature, thus making it resistant to single point failure or a simple combination of failures. This principle is illustrated in Fig. 6.4. Case examples of medical decision making in the face of byzantine faults is presented below.
Case Example 1
KK is a 39-year-old female referred for weakness in both hands. This has been present for at least 4–5 years but may have been present for longer. 4–5 years ago the patient had surgery on a ganglion cyst on her right wrist. After the surgery when her cast was removed the physician noticed weakness and wasting in both Abductor Pollicis Brevis muscles. Because of this she was referred to a neurologist who thought she may have a neuropathy as the cause. Eventually it was felt that she had a myopathy and needed evaluation at a university medical center. She feels the weakness is in her hands and distally in her legs. She has mild foot drop but this hasn’t caused any falls and she tries to remain active and athletic. The weakness is painless but now it is starting to cause some functional problems with opening jars. She feels that her right side is slightly worse than her left. She had a CK checked and it was in the 2000s. She has never had any change in her urine color. The weakness doesn’t fluctuate. She has not had any rashes or skin changes with this. She has no difficulties with raising her arms above her head, fixing her hair, walking up stairs, or rising from a chair. The patient is of English heritage, has a normal 16-year-old daughter and denies any similar illness or known neurological condition in her parents or siblings.
On examination no skin/nail changes characteristic of Dermatomyositis was seen. She had normal strength in her neck flexors and extensors, deltoids, biceps, and triceps. The right wrist extensors were 4+/5 and the left were similar. Wrist flexors were 5/5 bilaterally. Finger extensors were 4−/5 bilaterally. The flexor pollicis longus was 5−/5 bilaterally. The hand intrinsics showed severe weakness and wasting involving thenar eminences, hypothenar eminences, and FDI. Hip flexors showed only mild weakness being 5−/5. The quadriceps and hamstrings were normal. Bilateral tibialis anterior showed 4/4 strength with a mildly steppage gait. The medial gastrocnemius was normal showing 5/5 strength bilaterally. Deep tendon reflexes showed normal 2+ reflexes in the biceps, triceps, brachioradialis, knee, and ankle jerks. The sensory examination was normal. She was able to toe walk but unable to stand on her heels. The remainder of the clinical examination was normal. A review of records from outside showed high CK levels, ranging between 1,000 and 2,000 for a female weighing 55.8 kg or 123 lbs. The most recent CK measurement was 1962 performed in September 2012. Laboratory investigations for ANA, ANCA, SSA, SSB, HIV, and Syphilis were negative. Other data included HBA1c: 5.0, TSH 1.135 and CRP: 2.7. A NCS/EMG (shown in Table 6.1) was reported to be “consistent with a moderately severe distal myopathy with denervation potentials noted on EMG.” Based on the EMG findings, the right tibialis anterior was considered the best muscle to biopsy. The patient was steroid naïve at this point. An FTA was performed as shown in Fig. 6.5. The main conclusions of the FTA can be divided into two groups—Inflammatory Myopathies and Distal Myopathies [7, 8].
Table 6.1
NCS/EMG for case Example 1
Related posts:
Stay updated, free articles. Join our Telegram channel
Full access? Get Clinical Tree
