The severity of injury to the articular surface during an intraarticular fracture plays an important role in determining the risk of developing posttraumatic arthritis. Injury severity traditionally has been determined using largely qualitative and subjective means. Typically, clinicians use information present on radiographs in several ways. The most formal of these is to classify the fracture. However, classification is hampered by various problems including poor interobserver reliability and reproducibility,3,9 imprecise definition of groups,3 and lack of correlation with outcome.4 Furthermore, these techniques do not accurately reflect the energy of the injury.
Bony comminution is one indicator of the energy of the injury that can be determined quantitatively and on a continuous scale. This way of measuring injury severity fits well with fracture mechanics theory, which indicates that the greater the energy imparted in creating a fracture, the greater the liberated surface area.5,6 More comminuted fractures liberate greater fracture surface area compared with less comminuted fractures.
We have developed an image analysis algorithm to calculate liberated fracture surface area as an objective measure of fracture comminution. Information present in computerized tomograms is used for the required analysis. Proof of concept previously has been shown by using the algorithm on sections of the tibial diaphysis of bovine bones that were fractured with known energies.1 Clinical cases add several complexities to the measurements because of the greater biologic variability in the density of hard and soft tissues; the presence of cortical and cancellous bony regions; CT data that are not necessarily ideal or standardized; interfering metal from splints or temporary fixators; and in vivo bone water content. Moreover, the amount of energy or surface that differentiates fractures in humans (or required precision of the algorithm for clinical application) is not known. In this study, we examine whether it is feasible to use this algorithm to discriminate between clinical tibial plafond fractures.
MATERIALS AND METHODS
Fracture surface area was evaluated by one observer in two patients with tibial plafond fractures who were treated at our hospital. The Institutional Review Board granted permission for this study, and all patients signed informed consent.
Bony injury can be addressed in engineering terms, as a question of energy balance. When a crack propagates through a brittle solid, the energy driving that crack-propagating process is directly proportional to the free surface energy of the newly exposed area. Computed tomography provides the framework for the image analysis used to quantify the area. Using a modified region-growing approach, bony fragments in the data are delimited. Full details of this procedure have been described previously.1 First, bone fragments evident on the CT scan must be identified manually and seeded by the operator. During the region-growing process that follows, pixels that are similar in brightness and close in proximity to the operator-registered fragment seed are associated iteratively. Throughout the process, the criterion for brightness similarity is updated to reflect the brightness of the pixels that already have been defined as part of the fragment. The empirically optimal stopping point for designating the fragment border is that point at which an additional increment of the brightness similarity criterion will produce an abrupt increase in perceived object size. Once the bony region is grown, edge surfaces are identified; edge points then are ordered and measured. The image analysis of bone fragments incorporates the local density of the bone tissue (CT Hounsfield value) because this property is tied to the indigenous mechanical integrity of the bony material.
In determining the amount of surface that was newly generated by the injury, the original (intact) periosteal and endosteal bone surface area was determined from the contralateral limb. This intact area then is subtracted from the total measured area in the fractured limb. By scaling the liberated surface area by the free surface energy for bone, the absorbed energy can be estimated. The value of free surface energy (1.71 kJ/m2) used here was derived from an impact test done at 7.2 months/second on femoral bone from humans.2 The use of 1.71 kJ/m2 may represent a fairly liberal estimate of the energy absorption, because this was derived solely from cortical bone, and the surface area measured in the present cases includes cortical and cancellous bone.
A 59-year-old man fell 10 feet from a roof; he landed on both feet. He sustained a left tibial plafond fracture and minimally comminuted bilateral calcaneal fractures. His neurovascular examination was normal and he had mild swelling and ecchymosis over both legs without obvious deformity. Radiographs of the left ankle were obtained (Fig 1A–B), which revealed a relatively low-energy fracture of the tibial plafond. His foot was immobilized in a splint and a CT scan (voxel size 0.253 × 0.253 × 0.5 mm; Toshiba Aquilion scanner, Toshiba, Tustin, CA) of both lower extremities was obtained (Fig 1C). This was used as the source data for the described analysis. Six days after injury, he was treated with spanning external fixation and limited internal fixation of the articular surface with two screws.
The image analysis algorithm identified eight fragments. Two of those appeared partially attached to the primary (8878 mm2) fragment. The smallest fragment, a purely cortical bone chip, only was 31 mm2. One of the articular fragments, measuring 314 mm2, was impacted. Overall, 45% of the surface area was contributed by cancellous bone. The average Hounsfield value of pixels identified as cortical bone regions was 3.1 times the average Hounsfield value of cancellous bone pixels. Taking these two pieces of data together and scaling cortical and cancellous surfaces based on a power of density, approximately 9–16% of the total energy absorbed in this particular fracture was absorbed in cancellous bone. The total surface area on the intact right side measured 16,934 mm2. On the fractured side, surface area was 20,023 mm2, yielding a net liberated surface of 1545 mm2. The value is divided by two, because each crack produces two mating surfaces.
The measured area approximates energy absorption of 2.64 J in the left distal tibia. The available energy in the estimated 10 feet that the 80-kg man fell from the roof is 2.39 kJ. His impact velocity would have been approximately 7.73 m/second velocity = (2 × gravitational constant x height½). Therefore, the calculated 2.64 J indicates that less than 1% of the available energy was absorbed by the tibia in his left leg.
A 42-year-old, 109-kg man presented after falling 10 feet from a ladder. He had a 1.5-cm open wound over his distal fibula. Radiographs revealed a right tibial pilon fracture (Fig 2A–B). Initially, a spanning external fixator was applied. Five days after injury, his articular surface was reduced percutaneously and internally fixed with five screws. Computed tomography data from separate scans of each lower limb were reconstructed with a voxel size of 0.243 × 0.243 × 1 mm on the injured side, and 0.304 × 0.304 × 1 mm on the contralateral side (Fig 2C).
Sixty-eight fragments were measured in this comminuted fracture. They ranged in size from 10.2–22,400 mm2. In contrast to the first case, a significant portion of the total surface area came from small fragments. Approximately 51% of the surface was contributed by cancellous bone. The total surface area on the unfractured contralateral tibia was 23,886 mm2. The summed area of all the bony fragments on the fractured side was 74,940 mm2. The net liberated surface area is 25,527 mm2. The value is divided by two because each crack produces two mating surfaces. The approximate energy based on 1.71 kJ/m2, is 43.7 J. This value is more than 16 times higher than the energy estimate for Case 1. There also was a greater percentage of the total available energy, which was dissipated by the tibial fracture.
Many of the problems with classifying injury severity are explained by the concept that injury severity occurs on a continuum, yet classifications are categorical. There is inherent subjectivity in arbitrarily compartmentalizing a point in the continuous spectrum of comminution phenomena to the status of a categorical variable.9 Any divisions imposed in the continuous spectrum of fracture comminution severity ultimately are dependent on individual discretion. The current quantitative approach is not categorical. Likewise, it has shown merit as an indicator of energy absorption in our systematic, laboratory investigations.1
The data presented here are based on one assessment of each case by one observer; the intraobserver and interobserver repeatability of this technique applied to pilon fractures has not yet been documented. However, there was no significant difference between sets of measurements independently made by two operators of geometrically regular bone surrogate objects.1 Expected sources of intraobserver and interobserver variations are the operator’s ability to accurately identify all fragments and the selection of seed points. One study suggests3 that there can be observer difficulty in reliably identifying articular fragments on radiographs of plafond fractures. Strategies could be investigated for fragment identification without manual seeding, based on characteristic gray level intensity, gray level gradient, and/or fragment size and shape; the analysis and reanalysis of the same case then would always produce the same result.
One assumption made in our application of linear elastic fracture mechanics to bone fracture is that bone behaves similar to a brittle solid under impact conditions. Piekarski6 concluded that treating bone as a brittle solid was rather a simplification, at least for one crack propagating under quasistatic conditions. Bone is viscoelastic, and therefore exhibits strain-rate sensitivity. The assumption that is made is that under the regime of impact loading involved in the spectrum of pilon fractures, the strain rate does not vary considerably. Wright and Hayes10 have shown that an increase in strain rate of four orders of magnitude only produces a 75% increase in energy absorption.
Another assumption made in the analysis of clinical cases is that there was no preexisting damage or disease in the fractured tibia. Because fracture toughness parameters are material-dependent, surface area measurements incorporate the local material variability, available in the form of CT Hounsfield density. However, microcracks theoretically could diminish bone impact strength7 without having a significant effect on local Hounsfield signal intensity. This raises the issue that the resolution of the CT data can influence the results. The minimum necessary CT resolution for this analysis and the minimum detectable fragment size need to be defined more rigorously. In higher-energy injuries, smaller fragments tend to contribute a greater percentage of the total liberated surface area than they do in lower-energy injuries (based on our laboratory work in bovine bone).
It is important to realize that this technique will not be applicable to every pilon fracture. For example, the bony surface lost when there are missing fragments from a breach in the soft tissue cannot be estimated. Although we chose pilon fractures for analysis, the paradigm of inferring the energy of the injury from the degree of comminution is applicable to other fractures in different parts of the anatomy. Bone fracture energy is not necessarily tantamount to articular injury severity depending on the injury mechanism (for example, a direct impact in a vehicle-pedestrian accident).
Although the image analysis algorithm discussed here has been validated with in vitro laboratory tests, the cases presented are a first attempt in direct application to trauma in humans. Numerous issues still may be open to debate, including the relative consideration in the algorithm of cortical versus cancellous bone, the use of the contralateral limb surface (or implementation of a more practical allometric approximation) as an estimate of prefracture surface area, and the actual interpretation in energy terms of the surface area. However, although the absolute value of absorbed energy for these cases may be reinterpreted later, the relative comparison still is valid. Indeed, the output of the algorithm matched the judgment of the treating physician (JLM) as to which case represented the more severe injury and roughly approximated an estimate of the magnitude of difference of injury severity between the two cases. This can be judged independently by observing Figures 1 and 2. This difference between the cases here is such that the liberated surface area represents a 9.1% increase over the corresponding contralateral in Case 1 (near one end of the spectrum), compared with 107% of the contralateral in Case 2 (near the other end of the spectrum).
As the next step in validating this technique, a larger series of fractures needs to be assessed and ranked for injury severity using the algorithm and compared with experienced clinician judgment, because this is the only currently available gold standard. To be clinically applicable, the current procedure to quantify the liberated fracture surface area must be additionally automated. Currently, measures are being taken to increase the computational efficiency of the algorithm, so that it may be done quickly.
Our eventual goal is to develop a clinically applicable continuous measure of the energy transmitted through the articular surface during a fracture event that is reliable and reproducible. Such a measure, applied rigorously to a large cohort of patients, will aid in the development of optimal treatment procedures for particular levels of injury. It will help clinicians to think in terms of energy. Currently, we do not necessarily understand the implications of given energy delivered across the joint in vivo. Available estimates of articular cartilage damage from blunt impact have been derived from in vitro studies,8 and therefore are difficult to translate into any meaningful clinical guidelines. The technique that we have described has tremendous potential for clinical research because objectifying the severity of injury across large series of patients will allow us to begin to assess the effect of our treatments in ways currently not possible.
1. Beardsley CL, Bertsch CR, Marsh JL, Brown TD: Interfragmentary surface area as an index of comminution energy: Proof of concept in a bone fracture surrogate. J Biomech 35:331–338, 2002.
2. Currey JD: Changes in the impact energy absorption of bone with age. J Biomech 12:459–469, 1979.
3. Dirsch DR, Adams GL: A critical assessment of factors influencing reliability in the classification of fractures, using fractures of the tibial plafond as a model. J Orthop Trauma 11:476, 1997.
4. Gaston P, Will E, Elton RA, McQueen MM, Court-Brown CM: Fractures of the tibia: Can their outcome be predicted? J Bone Joint Surg 81B:71–76, 1999.
5. Griffith AA: The Theory of Rupture. In Bienzeno CB, Burgers JM (eds). Proceedings of the First International Congress for Applied Mechanics Delft, Holland, Waltman 55-63, 1924.
6. Piekarski K: Fracture of bone. J Appl Phys 41:215–223, 1970.
7. Reilly GC, Currey JD: The effects of damage and microcracking on the impact strength of bone. J Biomech 33:337–343, 2000.
8. Rohl L, Linde F, Odgaard A, Hvid I: Simultaneous measurement of stiffness and energy absorptive properties of articular cartilage and subchondral trabecular bone. Proc Inst Mech Eng [H] 211:257–264, 1997.
9. Swiontkowski MF, Sands AK, Agel J, et al: Interobserver variation in the AO/OTA fracture classification system for pilon fractures: Is there a problem? J Orthop Trauma 11:467–470, 1997.
10. Wright TM, Hayes WC: Tensile testing of bone over a wide range of strain rates: Effects of strain rate, microstructure and density. Med Biol Eng 14:671–680, 1976.