Introduction (Paul Dilley)
The Medinet Madi corpus of seven Coptic codices, containing translations of otherwise lost Manichaean writings, constitutes a key source for late antique religions. One of the Medinet Madi manuscripts without a critical edition is the so-called Synaxeis codex (CBL Pma 5), currently held at the Chester Beatty Library. Preliminary studies suggest that it contains selections (perhaps liturgical) of Mani’s Living Gospel, which had been almost entirely lost, except for a few scattered excerpts. The conserved pages of the Synaxeis codex will be published by Wolf-Peter Funk. Yet the pages of another section of the codex, the book-block, are fused so closely together from dampness that they are not separable without an unacceptably high risk of damage.
Given this condition, the Synaxeis book-block is an excellent candidate for X-ray tomography, a process that can identify the components of a book, even its internal text, without fundamentally altering its structure. This project was planned and carried out by a team composed of Professor Paul Dilley (University of Iowa), Professor Brent Seales (University of Kentucky; director, Digital Restoration Initiative), Christy Chapman (Research and Partnership Manager, DRI), Seth Parker (Technical Project Manager, DRI), and Kristine Rose-Beers, ACR (Head of Conservation, Chester Beatty Library). The Dumbarton Oaks Project Grant funded the preparation and transfer the book-block for X-ray imaging at Diamond Light Source (Oxfordshire, UK) on September 25–27, 2019.
Packing through Vacuum Encapsulation (Kristine Rose-Beers)
Due to the extremely fragile nature of CBL Pma 5, a packing solution was sought to mitigate the risk of damaging vibrations during transportation and enable the manuscript to be fully supported during X-ray tomography. Awareness of the conservation of a previously damaged manuscript, the Fadden More Psalter, a ca. 800 AD waterlogged parchment conglomerate manuscript at the National Museum of Ireland, by Dr. John Gillis, its senior conservator, led to the investigation of vacuum encapsulation as a possible solution.
To ensure the optimum stability of CBL Pma 5 within a vacuum-sealed enclosure, Escal® film, specifically designed for the long-term storage of museum objects in a vacuum environment, was used. The inner layer of Escal® is polypropylene, while the outer barrier layer is a vacuum-deposited ceramic on a PVA substrate. The film has oxygen permeability and water vapor transmission, keeping the enclosure content stable for at least 50 years. 30 gsm Bondina, a nonwoven polyester material with an extremely smooth surface, was selected as a support surface to minimize the risk of abrasion during transfer from the existing glass support sheet to the Escal® enclosure, and to provide a barrier between the manuscript and the support materials. 7 mm black Plastazote® LD45, a cross-linked closed-cell polyethylene nitrogen expanded foam, was used as a secondary support and cushion to disperse the vacuum pressure during enclosure. Due to the extremely friable nature of CBL Pma 5, a Milty 5036694022153 Zerostat 3 Anti-Static Gun was used to dissipate static.
CBL Pma 5 was encapsulated by Kristine Rose-Beers at the CBL on August 15, 2019. Once safely enclosed, a small identification mark of two hatched lines was drawn on the top right-hand corner of the sealed enclosure to mark the head edge of the object’s recto.
The final packaging allows the book-block to be turned, moved, and handled without direct contact or risk of loss. Although the manuscript is not visible, it is physically and chemically stable enough for both transportation and X-ray tomography. The enclosure is also completely reversible, and can be removed without detriment to the manuscript as necessary.
In order to maintain the integrity of the encapsulated manuscript, it was essential that the vacuum-sealed enclosure was not compromised at any time during transportation or imaging. As such, the manuscript was placed in a Plastazote®-lined phase-box in preparation for transportation to the synchrotron facility.
Mount Design (Digital Restoration Initiative)
Prior to the scan session, we identified two major risks to the manuscript, which our mount design addressed. First, the surface of the manuscript would be damaged by abrasive contact; this risk was addressed through the enclosure described in Part II. Second, the thin edges of the manuscript could not support the manuscript’s weight when placed in a vertical orientation, which is necessary for the CT setup at the Diamond beamline. To address this risk, we designed a 3-D–printed mounting enclosure that sandwiched the vacuum-sealed manuscript between additional layers of foam sheets and a rigid shell, minimizing pressure on the edges. Once assembled, the enclosure was placed into a circular base that could be secured to the Diamond sample table.
Tomographic reconstruction works best when the object being scanned has a relatively uniform shape while being rotated. X-rays passing through such an object will have a consistent attenuation once they reach the X-ray detector. Unfortunately, books and bound manuscripts do not have this uniformity. The large flat surface of a page is significantly wider than the total depth of the manuscript. As such, X-rays that are projected through the depth of the manuscript will be significantly brighter than those which pass through the edge-on orientation. Such a disparity in attenuation can significantly degrade the quality of the tomographic reconstruction.
To address this issue, our sample mount was designed so that the attenuation through the depth of the manuscript could be fine-tuned. We constructed two 3-D–printed, half-cylindrical shells which were placed into the circular base alongside the manuscript enclosure. Once assembled, the entire sample mount forms a cylinder with the manuscript at its center. During scanning we filled two hollow chambers adjacent to the manuscript enclosure with plain printer paper to achieve an acceptable level of attenuation disparity.
Scanning Information (Digital Restoration Initiative)
The manuscript was scanned at the Diamond Light Source I12 beamline on September 25–27, 2019, thanks to an institutional grant of six shifts of beam time. The I12 beam can be tuned to produce X-rays with energies in the range of 53–150 keV. For all experiments, we used the monochromatic X-ray beam mode, which produces X-rays at one specific energy level. This degree of specificity is not available in the conventional X-ray sources found in laboratory CT scanners, and it enables enhanced quantitative comparison between CT scans captured at different energy levels.
Additionally, all scans were captured with the large field-of-view (LFOV) image sensor. This sensor captures an area of 97 mm × 25 mm at 21.5 μm per pixel (approx. 1180 pixels-per-inch). Though this field-of-view is the largest that I12 provides in an image sensor, it is still significantly smaller than the mounted manuscript, which has a diameter of 240 mm and height of 325 mm. To address this, every full scan of the manuscript was actually composed of 40 subscans with overlapping fields-of-view: 2 horizontal scans by 20 vertical scans. These subscans were stitched together to produce a single tomographic dataset for the manuscript.
Analysis of CBL Pma 5 by Ira Rabin and her colleagues suggests that the manuscript is written in iron gall ink. To help us tune our scan settings and our analysis pipeline to this type of ink signal, we mounted an iron gall sample from papyrus produced by Paul Dilley, which was scanned alongside the manuscript in all primary experiments.
Three full scans of CBL Pma 5 were performed with different scan settings. The first of these was captured at 54 keV with a 0.13 s exposure time, a 0.04-degree step angle, and a 1 m sample-to-sensor distance. Lower X-ray energy levels are more likely to produce visible contrast for iron gall inks because the X-ray energies are closer to the k-edge of iron, i.e., the point at which X-ray attenuation is significantly more pronounced. The second CT scan was captured at 80 keV with a 0.035 s exposure time, 0.04-degree step angle, and a 2 m sample-to-sensor distance. This technique is called phase contrast tomography and is widely used to enhance subtle detail in scans of low-contrasting materials. The third CT scan was captured at 80 keV with a 0.035 s exposure time, 0.04-degree step angle, and a 1 m sample-to-sensor distance. This scan has the same sample-to-sensor distance as the first scan, but the same energy settings of the second scan, acting as a bridge between our previous experiments and creating a useful link for comparing regular tomography with phase contrast tomography for noninvasive text recovery.
Data Summary (Digital Restoration Initiative)
Scans were reconstructed with software internal to Diamond because the offset scanning method used required custom processing not yet available in an external software environment. The massive data sizes required several months of reconstruction, organization, and transfer time. In total, our scan session produced 15 TBs of raw and 5.3 TBs of reconstructed data, stretching the limits of our virtual unwrapping pipeline and storage system management.
Initial analysis of the internal structure of the manuscript was inconclusive for finding systematic ink signals. The reconstructed cross-sectional slices reveal an internal structure which appears to be significantly deteriorated, without obvious visible separation between pages. Using more advanced visualization tools will be crucial to understand better the structural characteristics of the manuscript.
The ink signal is not immediately apparent when observing the scan data directly. However, experiments on lab-produced samples indicate that the raw intensity signal of iron gall ink can be masked from direct observation by the highly irregular intensity levels of the papyrus surface. This masking effect would be especially pronounced for gall inks with low concentrations of iron. Recovering text from this data will depend upon the development of a local ink enhancement filter that can isolate and enhance any subtle ink effects that cannot currently be observed. We continue to work in earnest on a generalized approach for constructing such a filter from an extensive machine-learning reference library. We believe that this data will inform us as to the imaging tradeoffs between incident energies, resolution requirements, and overall virtual unwrapping success.
Next Steps
The goals of subsequent data processing are to determine the physical structure of the book block, in particular whether there are pages intact; and to detect and edit the surviving text within it. We envision several publications in both science and humanities journals describing our ongoing work.