*3.3. Lesional Analysis*

In group A, 155 lesions were pathologically verified prior to neo-adjuvant therapy and primary surgical treatment (breast: 86, locoregional lymph nodes: 58, distant: 11; Table S2A). Visual analysis of FDG PET correctly classified 115/155 (74.2%) lesions as malignant, and 16/155 (10.3%) as benign. FDG PET incorrectly categorized 24/155 (15.5%) lesions: 7/155 (4.5%) lesions in 7 patients were malignant but showed no uptake, whereas 17/155 (11.0%) lesions in 9 patients (5 with 1 lesion, 2 with 2 lesions, and 2 with 4 lesions) were benign but showed enhanced uptake (Figure 1). On this pathologically confirmed lesional basis, FDG PET had a sensitivity and specificity of 94.3% and 48.4%, respectively.

Group B consisted of 112 lesions (Table S2B). FDG PET classified 61/112 (54.5%) and 8/112 (7.2%) lesions as true positives and true negatives, respectively. Forty-three (43/112, 38.4%) lesions were classified incorrectly: 12/112 (10.8%) malignant lesions showed no uptake, whereas 31/112 (27.7%) lesions showed enhanced uptake reported as suspect but were benign. On this, with imaging/follow up confirmed lesional basis, FDG PET had a sensitivity and specificity of 83.6% and 20.5%, respectively.

Results of groups A and B taken together (Table S2C) yielded a sensitivity and specificity of 90.3% (95% CI 85.3–93.7%) and 33.3% (23.5–44.8%), respectively. Misclassification by FDG PET mostly involved axillary lymph nodes and bone tissue, respectively (Table S3).

**Figure 1. Examples of false negative and false positive lesions on FDG PET.** (**<sup>a</sup>**–**<sup>c</sup>**) Patient with primary ER+ breast cancer with faint uptake in the primary tumor (SUVmax 2.3). Low-dose CT (**a**) revealed a lytic lesion in the 10th thoracic vertebra (Th10) without enhanced FDG uptake (**b**). An MRI scan (**c**) revealed multiple vertebral metastases (Th4, Th11, Th12, L4, L5), including the one at Th10. This lesion was classified as false negative on FDG PET. (**d**–**<sup>e</sup>**) Patient with multiple mediastinal FDG avid, suspect lymph nodes. Coronal section of a low-dose CT-scan (**d**) and FDG PET scan (**e**). Endobronchial ultrasound–guided biopsy of 3 mediastinal lymph nodes showed reactive cells. These lesions were, therefore, classified as false positive on FDG PET.

A similar lesion-based analysis was performed for conventional imaging, including the diagnostic CT scan (Table S4), showing high sensitivity and low-moderate specificity rates of 95.9% and 15.2% and 80.8% and 66.7% for groups A and B, respectively. Outcomes of conventional and FDG PET imaging were also combined together for group A (Table S5), showing that conventional imaging alone identified 23 additional suspect lesions of which 7 were malignant. FDG PET alone identified 10 other suspect lesions of which 5 were malignant.

Quantification of visually identified lesions on FDG PET did not improve discrimination between true and false positives lesions (Figure 2, Table S6 and Figures S1–S3), in either group (A and B).

**Figure 2.** *Cont.*

**Figure 2. SUVmax and TLG show no significant differences between false and true positive lesions**. Lesions were classified into 3 groups, i.e. false negatives, false positives and true positives; lesions have been verified with pathology (**<sup>a</sup>**,**b**) or additional imaging and/or follow-up (**<sup>c</sup>**,**d**). Similar results have been obtained for SUVpeak and SUVmean (Supplementary Figures S1 and S2). \* *p* < 0.05.

#### *3.4. Correlation between FDG PET Parameters and Histopathology*

FDG uptake in the primary tumor was not associated with the accuracy of FDG PET staging (*p* = 0.67). Ductal carcinoma had a higher SUVpeak and SUVmean than lobular carcinoma (*p* < 0.05), and HER2+ tumors had a significantly higher TLG compared to HER2- tumors (*p* < 0.05) (Table S7). The % ER positivity correlated with TLG (*p* < 0.05)

#### *3.5. Implications for the Plan*

In summary, in 22/74 (29.8%) patients, the treatment plan based solely on FDG PET imaging would have been incorrect. In total, 65/74 (87.8%) patients underwent surgical resection, and in 34/65 patients (52.3%), surgery included also axillary lymph node dissection. No patient on neo-adjuvant therapy had progressive disease during treatment. Pathological analysis of axillary specimens classified 143 of 346 lymph nodes as malignant, whereas 203 were benign (Table S8). Since it is impossible to match each lymph node in the pathology specimen with their location on imaging, we compared the numbers of suspicious nodes on FDG PET with malignant nodes in the specimen. 83/143 (58.0%) malignant lymph nodes in 16 patients were classified as false negatives on FDG PET. In 7 patients, diagnosed with N1-stage disease on FDG PET, axillary lymph node dissection showed N2-disease (Table S9). In 2 patients with one malignant node on FDG PET, 1 or 2 additional nodes were identified when the axillary lymph node dissection was performed. Additionally, FDG PET falsely identified N3 disease (infraclavicular lymph node) in one patient, whereas in one case, N3 disease (intramammary lymph node) was missed. In the remaining patients, FDG PET showed the same number or fewer affected lymph nodes than the resection specimen, the latter most likely due to the effect of the neo-adjuvant systemic treatment. As the neo-adjuvant treatment affects the lesion size, no correlation between FDG PET positivity and the size of the lymph node metastasis could be made.

Metastatic disease was missed by FDG PET in two patients: one patient had multiple bone metastases and the other patient had a lung metastasis. In eight other patients, false positive lesions were identified in the liver, thyroid, bone and lymph nodes located in the neck, mediastinum and inguinal region (Table S3).
