T staging esophageal tumors with x rays

With histopathology results typically taking several days, the ability to stage tumors during interventions could provide a step change in various cancer interventions. X-ray technology has advanced significantly in recent years with the introduction of phase-based imaging methods. These have been adapted for use in standard labs rather than specialized facilities such as synchrotrons, and approaches that enable fast 3D scans with conventional x-ray sources have been developed. This opens the possibility to produce 3D images with enhanced soft tissue contrast at a level of detail comparable to histopathology, in times sufficiently short to be compatible with use during surgical interventions. In this paper we discuss the application of one such approach to human esophagi obtained from esophagectomy interventions. We demonstrate that the image quality is sufficiently high to enable tumor T staging based on the x-ray datasets alone. Alongside detection of involved margins with potentially life-saving implications, staging tumors intra-operatively has the potential to change patient pathways, facilitating optimization of therapeutic interventions during the procedure itself. Besides a prospective intra-operative use, the availability of high-quality 3D images of entire esophageal tumors can support histopathological characterization, from enabling “right slice first time” approaches to understanding the histopathology in the full 3D context of the surrounding tumor environment.


Table S1
Table matching the figures presented in the article with the esophageal specimens they were taken from, tumor stage at that specific position along the esophagus (corresponding to the presented CT slice), and ring removal procedure that was applied.All samples come from esophagectomies, meaning all patients were diagnosed with esophageal cancer; however, specimens typically also contained cancer-free regions due to lesions being localized.Similarly, different parts of the same specimen may contain lesions at different stages (see Fig. 6 in the main article and related discussion).The radiologists were expecting to find a tumor, and their task was to stage it.Four specimens were sufficient to provide a range of tumor stages, namely T1, T2, and T3; this was considered sufficient for a technical study aimed at obtaining proof-of-concept that esophageal tumors could be staged by means of X-ray images alone (as opposed to a clinical study).Instances of T4 cancer were not observed, which is not surprising as stage T4 patients are typically considered unsuitable for esophagectomy.h) and (j) for images in panels (a), (c), (e), (g) and (i), respectively).As can be seen, grey levels are not significantly affected by the angular down-sampling in the images; the grey level distribution in the histograms becomes increasingly smoothed out with reduced scan times, indicative of the reduced contrast-tonoise ratio.Although reasonable quality images are still obtained with datasets corresponding to scan times as short as 30 and possibly even 15 minutes, it should be noted that e.g. the 30 minute ones were obtained with 3,000 projections, which is already slightly below the recommended number of CT projections when a detector with 2,400 pixels is used.A correct optimization of scans performed in shorter times should entail collecting more projections with a shorter acquisition time per projection, which was made impossible by our protocol imposing us to perform a single, long scan for each sample (see main article).The color scale represents the attenuation coefficient µ eff in mm -1 (see equation ( 6) below), and the window width has been optimized to enhance the contrast of soft tissues.This has been applied in the same way to all presented images (which indeed all have a similar windowing), hence this is not repeated in the following captions.because tissue distortions associating with the embedding and cutting required by histology meant the lesion ended up outside the considered CT slide.Since similar lesions are regularly detected in all considered samples (with the lesion at 2 o'clock in the same sample providing a good example), we would be inclined to consider the latter explanation as more likely.In panel (c), a peripheral lymph node is highlighted by the radiologists; this was indeed an affected lymph node, as is clearly visible (and highlighted with a blue arrow by the pathologist) in the histology counterpart in panel (d).While on the one hand this hints at the possible inadequacy of T staging alone, it also suggests the possibility of using the technology for N staging, so long as a sufficient number of lymph nodes are collected and scanned -however, this laid beyond the scope (and ethics protocol) of the current work.
The color scale represents the attenuation coefficient µ eff in mm -1 (see caption of  visualized in (a).This is a similar situation to that observed in Fig. S1 (b), however it should be noted that the match between CT and histology slice is worse in this case, as it was more difficult to select exactly the same area from the x-ray dataset (e.g.see the differences in the lumen's shape, especially around 9 o'clock, and the significantly reduced area in between the two lumen "branches").In panel (e), the radiologists segmented three different regions, labelled as A, B and C. Region A corresponds to that highlighted by the pathologist with an arrow, as it is the one where the tumor invades into the muscle layer.The lesion of region B is clearly visible in panel (f), although it had not been highlighted by the pathologist since it does not invade into the outer layers.Finally lesion C is mostly missed as half of the histopathology slice was missing in this case.The color scale represents the attenuation coefficient µ eff in mm -1 (see caption of  Eq. ( 5) is only valid for monochromatic radiation.To apply it to a polychromatic source, one needs to 1) consider that the EI method is achromatic, a fully valid assumption if transmission through the masks can be neglected [2], which means C (and therefore C') do not depend on λ, and 2) by replacing µ and γ with their "effective" counterparts µ eff and γ eff , obtained as the spectrally weighted sum of all monochromatic components [1,3].This allows accessing the line integrals required for CT reconstruction by taking the logarithm of the inverse Fourier transform of the remaining terms:

Fig. S2 :
Fig. S2: Three more examples of stage T1 esophageal tumors, with XPCI CT slices (a, c, e) juxtaposed to their corresponding histopathological slices (b, d, f).In all cases, the tumor has been segmented by the radiologist in the CT slices, while the pathologist has put a red arrow in the histology slices indicating the point where the tumor has just started to invade the supportive layer that divides the submucosa and muscle layers, indicative of a T1 stage cancer.In panel (b), an additional arrow (at roughly 7 o'clock) highlights a second (T1) tumor formation which is not visible in the CT slice in (a), either because it was not detected, or Fig S1 for details).

Fig. S3 :
Fig. S3: Additional examples of stage T2 esophageal tumors, with XPCI CT slices (a, c, e) next to the corresponding histopathological slices (b, d, f).The segmented areas in (a) and (c) match the corresponding areas in (b) and (d), in which arrows drawn by the radiologist point at areas where the tumor is invading into the muscle layers, indicative of T2 cancer.The pathologist highlights an additional area in (b), corresponding to a lesion that is not Fig S1 for details).

Fig. S4 :
Fig. S4:Additional examples of stage T3 esophageal tumors, with XPCI CT slices (a, c, e, g, i) next to the corresponding histopathological slices (b, d, f, h, j).In all these examples, the tumor is very extensive and significantly disrupt the submucosa and both muscle layers invading into the adventitia (see red arrows by the pathologist in panels (b, d, f, h, j)), corresponding to T3 stage cancer; indeed, all these cases were stages ad T3 by both radiologists and pathologist.As mentioned in the main manuscript, specimens were too large to fit into a single histological slice, and as a result the slices shown in this article are obtained by pasting together two separate "half" slices.Because of this, half a slice is missing from panels (b), (h) and (j) -however the available half is sufficient to observe tumor invasion into the various layers.All cases are straight matches, with the pathologist using one or more arrows depending on the size of the "front" of the tumor invading into the adventitia.Panel (g) shows a case where the radiologists have segmented a smaller area, which again could be due to the worse match between CT and histological slices.Also in this case, however, they have staged the tumor as T3, in (blind) agreement with the pathologist.The color scale represents the attenuation coefficient µ eff in mm -1 (see caption of Fig S1 for details).